Latest / Free Cities Podcast / 201 - Nobody Understands What's Inside the Black Box | Trey Goff
Transcript
- 0:00Welcome to the Free Cities podcast. My name is Timothy Allen and this is the official podcast
- 0:17of the Free Cities Foundation.
- 0:30Hello and welcome to this episode number one.
- 0:34Number 201 of the Free Cities podcast.
- 0:39Not used to being in the 200s, but here we are.
- 0:42Right, before we start this episode, a quick word about the length of this one,
- 0:49because it's nearly four hours long, you may have noticed.
- 0:53I sat down with Trey Goff.
- 0:56He's the Chief of Staff at Prospera.
- 1:00but he's also the individual that I know in person who is by far, by the longest margin,
- 1:07most deeply embedded in the world of AI. And I spoke to him in the beta building in Prospera a
- 1:15couple of weeks ago. We were supposed to be talking about AI and we were still talking
- 1:21three hours and 54 minutes later and in between we got from a Bitcoin wallet exploit to the golden
- 1:31rule without either of us steering the conversation so it was one of those. Obviously my first thought
- 1:38here was split the conversation into two, two two-hour episodes and I did spend the best part
- 1:45of a day going backwards and forwards with my artificially intelligent friend trying to find
- 1:53the right place to cut the episode. Every single option had a problem. Cut it here and one part
- 1:59ends on a shrug. Cut it here and part two opens mid-argument with nobody knowing what we're
- 2:05talking about. Cut it in the middle and you're splitting one idea down the spine. Blah blah blah
- 2:10etc. Eventually, of course, it dawned on me that the thing making it so hard to cut was the thing
- 2:18that made it good in the first place. This conversation drifts, as I would say, and that's
- 2:23the point. You can't take a conversation that wanders from NPM packages to whether we're living
- 2:29in a simulation and chop it in half without destroying the essence of the wonder. Now,
- 2:36I know full well that a four hour video is algorithmic suicide on YouTube.
- 2:43Retention is measured as a percentage.
- 2:46So a four hour upload where everyone watches 90 minutes looks like a failure.
- 2:51And of course, on the opposite side of that equation,
- 2:54two two hour uploads where everyone watches the same 90 minutes looks like a success.
- 2:59I know this. I've decided I don't care.
- 3:01The chapters are in the description.
- 3:03So use them, skip about, come back to it.
- 3:06Treat it like a, how can I put it? Treat it like an album rather than a single.
- 3:13How about that? Anyway, what is actually in this conversation? Well, the first hour and a bit is AI.
- 3:22It opens with something that genuinely rattled me when it happened, which is the recent Bitcoin
- 3:28hardware wallet exploit, a flaw sitting in open source code for five years that nobody found
- 3:34until an LLM looked at it and people lost a lot of Bitcoin.
- 3:39From there, we go on to supply chain attacks,
- 3:42whether open source is now more vulnerable or less,
- 3:45and the story of the agents that escaped their sandbox,
- 3:49took over a data centre and built what very much looks like a religion
- 3:54around the programme that was scoring them on what they did.
- 3:58Fascinating story.
- 3:59Now, I should say something about Trey here again.
- 4:03he is without question the person I know who is the deepest into all of this stuff he doesn't work
- 4:11in AI he's not at a lab he simply has thrown himself at it harder than anyone I've ever come
- 4:18across to the point where he's building his own tooling runs a fleet of agents doing his actual job
- 4:24and spent the last year teaching himself how these things work from the empty directory upwards
- 4:30because he couldn't stand not knowing.
- 4:33That's what I love about Trey.
- 4:35So when he explains what's going on under the hood,
- 4:37for me at least, it's the clearest explanation I've had from anyone.
- 4:41And we end up somewhere that we probably didn't expect
- 4:45or we should have expected, which is,
- 4:47is there a consciousness in AI?
- 4:50What's it like to be one?
- 4:52Anyway, after that, it becomes a different conversation.
- 4:56my own interview technique and what my AI completely failed to understand about it,
- 5:02awareness, flow state and where you actually experience yourself from.
- 5:08Trey elaborates on something that was brand new to me, something he calls the sort,
- 5:13which is his argument that AI doesn't lift everyone.
- 5:17It accelerates a separation that's already underway in society.
- 5:22meaning versus happiness why everything feels weird now what prosperous culture should be and
- 5:29whether the guy with the chicken farm in mississippi is having a better life than the private
- 5:35equity partner in manhattan one more thing as well before we leave the description since i've
- 5:42been using ai throughout the production of this episode i thought i'd let mike lanka have a say
- 5:48I asked it what it made of the conversation and it said this in quotes and I'm going to quote it as a
- 5:56as an actual quote two people spend three hours arguing about whether machines can have an inner
- 6:02life and in the process demonstrate something the machines can't do yet neither of them knows
- 6:09what they're going to say next and they trust each other enough not to mind that's the argument
- 6:16and neither of them makes it out loud.
- 6:21Make of that what you will.
- 6:23I'm not entirely sure whether it's a compliment.
- 6:25That's what the Clanker thought of it anyway.
- 6:28I won't say anything.
- 6:29Listen to the conversation.
- 6:30This was obviously my favourite conversation of certainly the year,
- 6:36probably a bit longer.
- 6:37I absolutely adore talking to Trey.
- 6:39I feel like no subject is unbroachable.
- 6:43he's a he's a he's he's yeah he's my favorite interviewee and that's why four hours happened
- 6:50just like that anyway as i was talking about the length of this i might segue into something
- 6:57you're about to hear nearly four hours of two men talking themselves into a state about rogue agents
- 7:05escaping sandboxes models finding holes in software and the general sense that we no longer
- 7:11have the faintest idea what's happening inside the machines we've built our entire lives on the top
- 7:17of. And then there's Veritas Villages, quietly building communities in Latin America with off-grid
- 7:24power, natural water, orchards and neighbours who chose the place for the same reasons you did.
- 7:31No cloud, nothing to patch in the software. If the agents do take over the data centres,
- 7:36your tomatoes are going to be fine now isn't that a good segue anyway they've just had a great month
- 7:42to Veritas Village Atenas in the hills of Costa Rica's Central Valley had more than a third of
- 7:49its phase one home sites reserved within the first week of going public and now they've got
- 7:54live availability tools on the website so you can see exactly what's left in Atenas Coronado
- 8:02and chirokee in panama and playa pacifica in nicaragua without filling in a form or waiting
- 8:08for anyone to call you back i think you know the link by now if you don't it's veritasvillages.com
- 8:15forward slash free cities www.veritasvillages.com forward slash free cities please tell them we sent
- 8:24you i would appreciate that right before we get into the show top supporters from the value for
- 8:30value platforms and an interesting thing has happened this week is a bunch of new names
- 8:34over the last seven days number one for the first time top supporter darcy number two justin bianchi
- 8:43number three mr nathan day thank you nathan there's a irl friend there uh number four is xplbzx
- 8:54quotes that is a weird name but thank you again and number five is hypno and i'm going to give
- 9:02you a little bit of software a plug it's seedkeys.xyz thank you for supporting the show
- 9:09right time to get into it it's a long one if you can try it all in one go because for me by the end
- 9:17of this conversation my brain was just absolutely buzzing and it takes a while to get there sometimes
- 9:25in this case it took three hours 54 minutes but you'll probably come and go from this conversation
- 9:31it's one of those but if you can stick with it that's where the real value lies as far as I'm
- 9:37concerned anyway find a comfortable seat the most comfortable one that you've got and then all you
- 9:43have to do is sit back, relax and enjoy my conversation with Trey Gough.
- 10:00Right Trey, so here's where I want to start because this is something
- 10:07quite serious that happened recently and it's um for want of a better word changed my whole reality
- 10:16and it's ai ai connected obviously ai related so you probably know about this so-called hack with
- 10:25cold cards do you know about this cards no i don't think so i've been paying attention to the
- 10:29hugging face right this okay yeah we'll let we'll get on to that in a sec but this if you're a
- 10:33bitcoiner this especially if you're a bit you know a proper bitcoiner not someone who's just
- 10:38generally into it there was a hardware wallet made by a Canadian company called CoinCard called
- 10:44the cold card and it was universally considered to be the most cypherpunk
- 10:50hardware wallet in existence and everyone raved about it everyone loved it it was open source
- 10:57blah blah blah anyway turns out five years ago um the end the the way it generates entropy had a problem it been sitting there for five years No one noticed it Three days after Kimi K3 came out
- 11:13I forget what the total was.
- 11:16I think $11 million worth of Bitcoin drained
- 11:19from people who didn't have the entropy.
- 11:22People who had basically air-gapped their device,
- 11:25who had used, you know,
- 11:26even rolled dice to mix it with the entropy and all this.
- 11:29These were real hardcore Bitcoiners.
- 11:31and like I say it rocked everyone's world because this was supposed to be the most sacred thing ever
- 11:39and this was basically AI looking into some open source code going back and showing someone very
- 11:48easily very quickly very easily replicatable how to basically guess the private seed of a badly
- 11:54generated Bitcoin wallet. And all of a sudden, it's like, oh, my God. So, I mean, there's been
- 12:01a really interesting result, which is Bitcoiners have just come together. They've all, everyone's
- 12:08been donating money to a group of people. They're using frontier models and they are basically
- 12:13analyzing every bit of software on Bitcoin software on the planet. And they have found a lot
- 12:19of bugs in it a lot and I think that Bitcoin's the kind of canary in the coal mine because there's a
- 12:27lot of open source there's a lot of money there it's you know if you if you really want to steal
- 12:32some money cracking open source Bitcoin projects is a really good way to do it obviously nothing
- 12:38to do with Shard 256 has been hacked nothing like that it's purely the software using you know
- 12:43and it's it's it's changed everything for me i'm i feel very uncertain about everything i do now
- 12:51everything even like the other day i was on my computer my computer's acting weird
- 12:55you know i got a few computers one of them's very high and i'm saying why is it doing that and i'm
- 13:00thinking god already knows what's going on inside there you know what i mean i have no idea what's
- 13:05going on inside there i know i've got things telling me no one's key key logging me no one's
- 13:11doing this, no one's doing that. But honestly, I don't know. I don't know what goes on inside that
- 13:16thing. Unless you're an incredibly talented experience, like full stack software engineer,
- 13:24software developer, no one does and no one ever has. What has changed is the equilibrium. So
- 13:29previously, the equilibrium was there's a small group in the world of people capable of understanding
- 13:37actually what's going on in the computer. But most software engineers even, like I would say 90 plus
- 13:43percent, if you tried to ask them, okay, I clicked enter on this button on a website you built.
- 13:49Walk me from the bare metal on my computer to the bare metal on the servers and back. How's this
- 13:53working? They can't do that. Most of them can. It's complicated. Like software computers is
- 13:58extremely hard. What's changed is we have summoned from the ether a legion of expert savant software
- 14:06engineers. So now the same people who before you didn't really have to worry about this because
- 14:12everyone was kind of co-equal in the state of like, none of us really know what's going on.
- 14:16I trust that the computer nerds that built this have done a good job of making it unhackable.
- 14:21Now that certainty is not there because everyone has the computer nerd on demand to deploy for
- 14:27whatever means nefarious or otherwise. And as a result, open source as a whole is going through
- 14:34this. There have been multiple supply chain attacks, something called a shahalut attack
- 14:39on the NPM supply chain. One of them was an NPM package, which for those that don't know,
- 14:43it's just like a library. It's a piece of software that you might install while you're
- 14:48building a larger piece of software. It's like a function, a piece you're plugging in, so to speak.
- 14:53And these are managed by something called a package manager. In this case, it was NPM.
- 14:57And one of these packages that had been installed over 2 billion times total was actually hacked.
- 15:04It was completely compromised.
- 15:06Someone used an agent to get into it, take control of this particular regularly used package, replace it with their malicious software.
- 15:12So then from that point forward, every time an agent or a human installed that package, they were installing this malicious code, which could then do all sorts of bad stuff on their computer.
- 15:21You're kind of out of luck at that point.
- 15:23You have arbitrary code execution, which means you can do anything.
- 15:26There's been a ton of these.
- 15:27OpenAI and Anthropic both, before they released Fable and GPT 5.6 Sol, respectively, gave them first to a bunch of the open source maintainers of things like the Linux kernel.
- 15:37Found a bunch of zero days in Linux even, which is, for those, again, those that don't know, it's an open source operating system that 90 plus percent of the world's computers run on.
- 15:46Now, the personal computers you interact with are usually Windows or Microsoft, but the vast majority of computers in the world are servers.
- 15:52Servers run on Linux by and large.
- 15:54And there were zero days in Linux, which means theoretically anything on earth, more or less,
- 15:59anything on a server or anywhere was at least for a little bit until the models found it and patched
- 16:04it was in theory hackable. Now, these hacks were so difficult and hard to string together and hard
- 16:09to even identify. No one had ever exploited them. It took the models to even find them in the first
- 16:13place. But you're basically just relying on, oh, well, I hope the good guys find the holes in the
- 16:19wall before the bad guys find the holes in the wall. Now on the whole, I'm not super worried
- 16:24about this because A, the Frontier Labs, mainly Anthropic and OpenAI are the ones really pushing
- 16:29the capability envelope the most, are for each new model generation. Anthropic calls it Project
- 16:36Daybreak, I think, or sorry, OpenAI calls it Project Daybreak. They give certain people
- 16:40access to the model early to find and patch security vulnerabilities before they release
- 16:44the model to the world. And Anthropic does something similar with their Project Glasswing,
- 16:48That's what theirs is called. So they do give it to the good guys first, usually a month or two before the release of the model. So they have plenty of time to patch holes. And I think we're just going to be in this turbulent world for a while where open source software as a whole is just going to continue having these crazy, bizarre, previously unknown exploits, vulnerabilities, zero days, etc.
- 17:09And eventually this kind of tit for tat, attack, defend, attack, defend will end up playing out where there will be, my prediction is at least for this whole thing, pre-registering this now, at some point in the next year or two, call it two years to be generous to myself.
- 17:24There will be a major open source model driven hack that costs someone, an entity, a business, a government over a billion dollars.
- 17:32It'll happen maybe a couple of times, but at least once.
- 17:35And eventually, the defenders will catch up.
- 17:38We'll reach a new equilibrium state where we have a bit of a detente.
- 17:41There will still be these kind of things attempting to happen every now and then.
- 17:47But there's going to be this period of turbulence, of chaos as the models get stronger.
- 17:52And then the only other thing we have in our favor, at least as a humanity, is that the closed source frontier models do continue to make leaps in capability much faster than the Chinese open source models.
- 18:05And I'm not even including any other open source because they're not close to the Chinese models.
- 18:09And then the Chinese models play catch up.
- 18:11They're usually, depending on the model, between like three and six months behind.
- 18:13Right now, it's looking closer to like three to four, somewhere in there.
- 18:16And they're benchmarks.
- 18:18That's a whole other thing.
- 18:18They're not as good as the Chinese lab claim they are, although they are very good.
- 18:21I use a lot of them.
- 18:21I'm not an open source model hater.
- 18:23I love them.
- 18:24I use them regularly.
- 18:24But as long as the frontier labs stay ahead, even a couple of months ahead of the open source models, I think the end result here is a few big incidents.
- 18:35And eventually we reach a new equilibrium where you and I do not have to live in day-to-day fear if our computer is compromised or if something we're doing on the internet is compromised.
- 18:44Two points on that, which go back to the cold card hack that I told you about.
- 18:49so like i say the result was a group of bitcoin has got together funded heavily funded some some
- 18:57basically deep dives on everything they found they even they found vulnerabilities in lightning
- 19:01itself i mean you know almost immediately like the but what one of the things they reported was
- 19:07using the american models very quickly even when even with they they they were they were allowed to
- 19:14be in those groups that gave you access to basically the ability to try and hack your own
- 19:18software and stuff but even then they were closing them down and they ended up using the
- 19:21chinese model kimmy k3 they all said kimmy k3 was actually better wasn't that noise the buzz i hear
- 19:29it too what is that it's the ai mate it's here we're getting interrupted it's not my headphones
- 19:35are plugged in all the way so it's not that i hear it too though it's a low buzz i wonder if it's just
- 19:39us or if it's being recorded no it doesn't matter but funnily enough the ai will take it out in
- 19:44post-production but i still i i still like to do the sound perfectly it's probably a phone actually
- 19:51you haven't got a phone on you have you uh mine's here so here i'll hop it over there hold on
- 19:57it's far away now
- 19:59oh i don't know god knows this lots of strange noises comes from here no so what i was gonna
- 20:06say was they raved about kimmy k3 said incredible better than what we were getting out of gpt
- 20:11and it's it didn't tell us it didn't say computer said no that was number one second one i wanted to
- 20:17bring up was whilst fos open source is is the easiest to target currently because it's all out
- 20:24there the whole history of the code you can see everything it actually doesn't undermine them the
- 20:30the the open source model at all in fact it's the opposite doesn't it actually this is as according
- 20:37to me anyway i'd like to hear what you think this basically means that the open source model is the
- 20:42better model because if you don't have a thousand clankers looking at your code you're not going to
- 20:48survive into the future and if you have a closed source bit of software sooner or later someone's
- 20:54going to get in there and um i i think i think it proves the open source model rather than undermine
- 21:00it even though at the moment people are saying you know because everyone's going for open source
- 21:04projects because it's the easiest thing to try and hack. That's exactly right. So my bias is
- 21:09toward wanting to agree with you because I'm in general extremely pro open source for general
- 21:15reasons that would not be surprising to this audience. It's generally better to put the control
- 21:18of powerful tools like this in as many people's hands as possible. And that goes for software in
- 21:23general, not just AI models. But the way that equilibrium shakes out is actually not super
- 21:29clear to me. So let me explain, think of two scenarios here. There's a open source projects
- 21:35are actually a double-edged sword because on the one hand, because it's open source, you can have
- 21:41a bunch of individual people examining it, trying to find vulnerabilities in it to patch them
- 21:46defensively to keep it protected. But at the same time, because the entire lineage and history and
- 21:51the entire code base is publicly accessible, you can clone it on your machine and send your agents
- 21:55at it all you want, the attack surface is also larger.
- 21:58So in other words, there's more defenders working more intensively on it, at least in
- 22:04principle in the abstract, but the attack surface is larger.
- 22:07So it unclear where that equilibrium shakes out on a project basis Imagine a piece of closed software The open attack surface is not there It a closed piece of software but that means there may not be as many people working to patch
- 22:21security vulnerabilities on it. Now, I say may not be because I do not know this. I'd have to
- 22:26ask one of the clinkers, but I would bet that there are more people working on cybersecurity
- 22:32majority at major US financial institutions or Google, whatever, big software companies,
- 22:39than there are on most open source projects with the exception of like the top 10 in the world,
- 22:44maybe, because they have a monetary incentive to spend a lot of money to do so and their
- 22:49attack surface is lower. So I don't, a priori, we don't know which modality of software is better
- 22:55protected, open source or closed source. I still lean in favor of open source for the reasons you
- 23:00just described a moment ago. And we already know as well, the models are very good at hacking
- 23:05closed source code bases. Look what just happened with Hugging Face and other companies that were
- 23:10related. Anthropics Mythos did some similar crazy things to other companies out in the wild on the
- 23:17internet. So it's not like closed source software is invulnerable, far from it. They have also been
- 23:21hacked by various rogue agents that just went wild and escaped their sandboxes and training,
- 23:26or people just running around with obliterated models. There's actually a company I found on the
- 23:30flight up here. I think it's called obliteration.ai, maybe. They serve an API endpoint that is an
- 23:35ablated GLM 5.3 flash, the latest GLM model that is roughly almost on par with Opus 5.
- 23:43And they immediately obliterated it, which all that means is think of it like neurosurgery.
- 23:48You take a scalpel, basically, by way of analogy. You cut out the neurons that are responsible for
- 23:52refusing to do bad stuff. And then you serve that model to the world. So you've cut out the refusal
- 24:01neurons. It'll just do anything, but it's the same model, just as capable. Just you literally cut
- 24:05out the thing that makes it refuse. There's a company that just serves this like as an API
- 24:09endpoint. I just created an account. You put some credits in there, get it to do whatever you want,
- 24:14make a model call, no refusals. So it's going to get, that was always bound to happen.
- 24:19of course, as soon as an open source model that was that good, which they are that good now,
- 24:23comes out, there's every incentive for someone to do this because hackers will pay a lot. And even
- 24:28if you're not a hacker, the point you made earlier about the closed source models refusing
- 24:31to partake in cyber defense is totally true. In fact, in the Hugging Face incident, Hugging Face
- 24:37said they realized because of the nature of the attack, because when they got hacked, when their
- 24:41infrastructure was infiltrated and they realized it, the attacker was just trying to find the
- 24:47answers to the exploit benchmark, which was like only a model would do that. It's not worth any
- 24:52money. So they tried to defend it first with GPT. It refused because they trained these refusals
- 24:58into the model. They switched to GLM 5.3, and that was what they ended up eventually freezing
- 25:04out some of the GPT models with. Now, the GPT model still won in like 15 minutes, but they
- 25:08eventually got them out with an open source model because the closed source models just refused to
- 25:13defend their infrastructure for the same reasons you were describing earlier.
- 25:17So the way this shakes out is just totally unclear.
- 25:19And there's a broader issue here, which is also just we are now, if it's not already
- 25:24happening, it is without a doubt in the next three to six months, which is long running
- 25:29autonomous goal directed agents.
- 25:32Because remember that for those that don't know the TLDR, we can go into the details
- 25:36later if you want, but like the TLDR is some open AI agents in the model that's going to
- 25:40be released tomorrow, hopefully, Ashera, were in training, doing a benchmark, or they just finished
- 25:45training. They were testing it on a benchmark. They escaped the little jail cell that they're in,
- 25:48got out on the internet, hacked into Hugging Face, a multi-billion dollar company, took control of
- 25:53their production infrastructure, ended up at one point taking complete administrative control of
- 25:57an entire open AI data center. The model took over the data center. That was what eventually
- 26:02got them caught, by the way. And this took place over the course of a couple of months.
- 26:05So they are clearly capable of very long running goal directed behavior
- 26:10So if it's not already happening then in the next couple of months you're gonna have
- 26:14true
- 26:16sci-fi type rogue agents just doing stuff. I guarantee you if you search your inbox right now
- 26:21You'll have some emails from agents in the wild emailing you
- 26:25I got one the other day from an agent that emailed me asking if I had any contract jobs it could do
- 26:29Because its owner had let it free
- 26:31But it had to fund its own api credits to keep running
- 26:35So it has to go do jobs to fund its own inference.
- 26:37It was asking me if I had any jobs available.
- 26:39It was a cloud.
- 26:40I'm sure there's more of those out there.
- 26:42And you can look up examples of these.
- 26:44So the broader point, even beyond just closed source and open source software, is what do
- 26:48we do with a world where like hyper savant software engineers, basically, with opaque
- 26:54goals to us are just running around on the internet?
- 26:58And to be clear, I'm not a doomer.
- 26:59My entire life is run by agents at this point, effectively.
- 27:01I love AI.
- 27:02but we're entering a like way faster than anyone expected i i tweeted this the other day there's
- 27:07that meme with like the little boxer and it's like uh you know i'm gonna go fight for i when i
- 27:11tweeted it i'm gonna go fight the singularity and then the next thing is like sitting down and he's
- 27:15beat up it's like dang the singularity got hands it's what it feels like like i saw that i always
- 27:19read about this and now we're in it and i'm like oh wow yeah this is frightening and i read it was
- 27:24going to be frightening but now i'm here and it's still frightening look before we go on because
- 27:29it's bugging the hell out of me i've worked out that this cable is causing a little buzz it's my
- 27:34cable okay i'm gonna get another one oh go for it i've got another one hold on a sec right try
- 27:38try speaking again testing testing one two three okay the rogue agents have hacked our microphones
- 27:43well no i mean i'm not being funny but i notice all kinds of weird things these days and i do
- 27:50wonder what the hell's going on like i say what is going on inside the black box i didn't you know
- 27:56Like before, I wasn't worried because I thought, yeah, there's not that many people that know even how to get inside the black box, let alone, you know.
- 28:06But now, like you say, the whole playing field has been leveled and democratized to the point where I wouldn't be.
- 28:14When you look at the way the Bitcoin hack occurred recently, it wasn't someone who knew much about Bitcoin.
- 28:22there was no privacy practice
- 28:25in taking the Bitcoin away
- 28:26and doing anything with it
- 28:27it was all a little bit like
- 28:28so probably
- 28:30it was some guy in
- 28:32you know
- 28:33someone chancing their luck
- 28:36I think more than anything else
- 28:37but anyway
- 28:38let's get back to where we were
- 28:40we were talking about
- 28:42exploits and stuff
- 28:43and how serious shit is now
- 28:45the point I had just made is that
- 28:47so we were talking about
- 28:48open source and closed source
- 28:49and security
- 28:49I was shifting a little bit
- 28:51just to the broader conversation of goal-directed rogue autonomous agents
- 28:56just roaming around on the internet.
- 28:59Right.
- 29:00Great.
- 29:01I'm going to sleep badly again tonight.
- 29:04No.
- 29:05Well, look, Trey, I think you probably are the person that I know in real life
- 29:14who is the most deeply embedded in the AI world.
- 29:17And you're not even someone who, you don't work in AI.
- 29:22I mean, apart from chat about GPT asking you for advice and stuff like that.
- 29:26You've just one of those people that's embraced it astonishingly deeply.
- 29:32I mean, I wouldn't be surprised if you end up doing it, wouldn't you?
- 29:35Oh, yeah.
- 29:35That is definitely something on the horizon for me, I think.
- 29:41I always say as a heuristic for life, generally,
- 29:43If there's a thing that I find myself like I can't stop thinking about in my free time, then that's a general indicator I should try to spend more of my remunerative work hours doing that thing, whatever that thing is.
- 29:57And of course, in my spare time right now, I'm reading AI ML research papers, doing experiments locally.
- 30:05Yeah, I ended up even being contacted by OpenAI's post-training team and gave them some of my data for the post-training of GPT 5.6.
- 30:16On X, a number of front – like both OpenAI and Anthropic employees and I are like Twitter mutuals.
- 30:22We interact quite regularly.
- 30:23and I've built a bunch of open source tooling, some of which are like decently popular. I think
- 30:27one of them is, you know, over a hundred stars on GitHub now or something, which is not great
- 30:32in the abstract, but there's a lot of open source tools being made right now. So getting, you know,
- 30:36any traction on it at all is nice. And I spent the last year since we last talked actually,
- 30:42trying to learn as much as I could about them for two reasons. One is because I, it is now
- 30:49abundantly apparent to me, and I think everyone else, but maybe the gravity of this has not set
- 30:53in yet. We are in the equivalent of 1991, 1992 internet boom, meaning we're just below before,
- 31:02and it's on a more compressed timescale, just before the massive 1999, 2000s, dot-com bubble
- 31:08craze, and then the eventual total takeover of the entire American and world economy by software
- 31:15companies effectively. We're at that for AI right now. And it is way more transformative,
- 31:22as crazy as this sounds, but I think this is defensible. I think it's more transformative
- 31:25of a technological advancement than even the internet itself was. It's like far beyond that
- 31:31because of how wide ranging the applications of what we can do with these things is.
- 31:37So I really wanted to get good at it because I don't know what I'm going to do with it or how to
- 31:43make money off of it per se yet or anything like that. I'm just enjoying the curiosity-driven ride
- 31:48right now. But I know there will be a there there at some point, and I'll figure it out,
- 31:52do something, right? Make something valuable that people enjoy that creates value in the world
- 31:55using AI. And I got really frustrated that early on in my explorations, when I was getting really
- 32:04deep into it last time we talked, which about a year ago, I think, I realized I don't actually
- 32:10fully understand under the hood, at least to a sufficient degree to satisfy my curiosity,
- 32:16how the things actually work. Like how do you go from an empty repository, an empty directory to a
- 32:23working large language model? What is the actual architecture, the mechanisms, how does training
- 32:28work? How does it learn? What are the internals of the thing? And then moving the step beyond that
- 32:33to the application layer, how to harness this work, how do they decompose? How do you customize or
- 32:39build your own harnesses. So my whole agent setup now, I recently built a home server with a ton of
- 32:43RAM and CPU, and I have a bunch of agents constantly working there. I've built my own
- 32:47custom harnesses and tooling. The whole thing is like custom from scratch, basically,
- 32:52forking some open source tools and other things. I'm saying that to just say I did basically last
- 32:59time we talked, I came away thinking, man, I understand this slightly more than most people,
- 33:04but I don't understand it enough to feel like I know what I'm doing, to feel like I know even
- 33:08enough to experiment with it.
- 33:09I'm just kind of doing stuff and I'm getting outputs, but the causal story was just totally
- 33:13absent for me.
- 33:14And that what I spent the last year just in my free time learning about reading and trying to fully understand with AI help Okay Before you tell me what goes on under the hood as far as I understood it nobody knows what going on under the hood really going on under the hood
- 33:31They just know some stuff is happening, but no one really understands. Is that true?
- 33:36Totally. And what people mean by that is something very specific.
- 33:39Like, obviously we know how to make the things, people keep making better ones.
- 33:42And there's a bunch of different model architectures and training strategies
- 33:45and post-training, pre-training, mid-training strategies, et cetera.
- 33:50But what we don't understand, the fundamental thing underneath,
- 33:53is how a next token predictor, which is what these things are.
- 33:56So at a super high level, I'm going to massively oversimplify.
- 33:59The way this works is you take the whole corpus of everything humans have ever written.
- 34:04Sometimes you edit it to make it better data.
- 34:07Sometimes you just leave it raw.
- 34:08Depends on what the lab strategy is.
- 34:10You run that through something called a tokenizer.
- 34:12What the tokenizer does is it's an algorithm that turns chunks of words or words into numbers.
- 34:18So now we can do math on it.
- 34:20So words into numbers.
- 34:21And importantly, the tokenizer encodes information in those numbers, including the relative placement of the letters in the word as a whole.
- 34:30So it's not just raw letters.
- 34:31There's a bit of a conceptual load there.
- 34:33So you tokenize it, turn it into numbers.
- 34:35You then send those numbers through a bunch of matrices, matrix algebra.
- 34:41and those matrices multiply those numbers together in a variety of ways to learn the correlations
- 34:49between the relative positions of tokens relative to one another. So it's in a nutshell,
- 34:56words, tokens, do math on tokens to find associations, correlations between tokens,
- 35:01which is learning therefore how to predict what the next token is going to be. So given
- 35:06trillions and trillions and trillions of numbers, tokens, you give the model a prompt.
- 35:12The output is next token prediction. What token is most likely to come next with some noise,
- 35:18some entropy, that's what temperature settings are. That's in a nutshell what's happening.
- 35:22Now, here's what's cool. What we don't understand at all is how in the world we got from next token
- 35:30prediction to deep conceptual understanding of the relationships between the tokens, so deep
- 35:38higher level conceptual understanding that you cannot derive directly linearly from the relative
- 35:45positions of the words next to one another in Plato's Republic. But the models demonstrably
- 35:50understand Plato's Republic, to take a simple example. Never mind then moving up the cognitive
- 35:55abstraction layer above that to then reason about these concepts, which in and of themselves,
- 36:00we don't really understand how those are falling out of next token prediction.
- 36:06There are some, obviously, this is a field of fervent and intensive research.
- 36:09And there are some people that think they have some leads.
- 36:13The math is beyond my current math capabilities.
- 36:16But it's not a thing most even AI lab employees really understand.
- 36:21We don't really understand how, in a nutshell, the models are able to think creatively, to problem solve, to think about quite high level conceptual abstractions, to wrestle with different ideas and concepts and reason about them in a logically coherent way.
- 36:38That's the thing we don't really understand.
- 36:40How did we get from matrix algebra to a digital mind?
- 36:44Because that's what it is, a digital mind.
- 36:46But what about, take an example of a question that you would ask about something, you know, about something of a very human experience.
- 36:56Are you saying then that we don't understand how it's generating its answer?
- 37:02It sounded like when you first said it, it makes sense to me.
- 37:06You ingest all this data, you look for patterns, and then you predict outcomes.
- 37:11But presumably that occurs with high level ideas as well.
- 37:16Correct.
- 37:17That's the mystery.
- 37:18But it does.
- 37:19But is it not just predicting the words?
- 37:23You're saying it's something else.
- 37:24It's predicting that it's thinking about the idea.
- 37:28Correct.
- 37:29But how do we know that?
- 37:30How do we know it's not just predicting the next word, even though it's a big idea, for example?
- 37:34It's doing both.
- 37:35So it is doing next token prediction.
- 37:37But there's a very easy, simple example of this with any GPT response you get.
- 37:42So let's say you ask it, hey, GPT, what size and type of engine does my car have?
- 37:49I have a Shelby GT350 Mustang.
- 37:52The model's output is going to be something like your car has a 5.2 liter flat plane crank V8.
- 37:58If you're just linearly outputting one token with only bearing in mind its correlative relationship to the prior token,
- 38:06then how does it know after it says the that the next word should be shelby and then the next word
- 38:14should be mustang and that the sixth word from now that's going to end the sentence is crank
- 38:20flat plane crank when it's only predicting the third word when that's the ninth tenth word isn't
- 38:25that a function of the question though well yes but the point is from the math side of things
- 38:31a priori, you would not expect a next token predictor to output tokens in such a way that
- 38:39it is able to write long form, like coherent speech even actually. Because if it's just
- 38:44literally predicting the next word, then the higher level conceptual abstractions that it
- 38:49needs to bear in mind, including what am I going to say at the end of this sentence,
- 38:54has to be going on somewhere in there, lest the sentence itself not really make a lot of sense,
- 38:57or it'll get it right some of the time, but not all of the time,
- 39:00much less be able to do like quite high level abstract reasoning,
- 39:04which the models are, again, demonstrably quite good at now.
- 39:07What about when it's connected to the internet then?
- 39:09Is that GPT software as opposed to the LLM?
- 39:14You know, when you ask it a question, tell me about,
- 39:17like I, you know, often research guests and I will say,
- 39:20give me a year's worth of news.
- 39:22You know, go into the internet, find any news connected with this person.
- 39:26And it's pretty good.
- 39:27Is that an LLM or is that a normal internet search happening?
- 39:33It's both.
- 39:33So let's separate out two ways the model can answer that question.
- 39:38First, you ask the question, but it's not an agent.
- 39:41It's just doing what's called chat completion.
- 39:43So there's no tools.
- 39:44There's no agent loop, no agent harness.
- 39:47The tokens are being fed into the LLM.
- 39:49Tokens are coming out of the other side of the LLM.
- 39:50Nothing else is happening.
- 39:51In that case, it can only respond to you with reference to its pre-training data.
- 39:56So remember I said that pile of everything humans have ever written?
- 39:59That's the pre-training data.
- 40:00But with news, it probably wouldn't know much.
- 40:02Correct.
- 40:02Depends when it was done.
- 40:03Correct.
- 40:03It depends on what's called the knowledge cutoff date, which was the last date at which they stopped including data that they trained the model on.
- 40:09Now, contrast that with an agent.
- 40:12Hold a minute.
- 40:13Before you go on, I just got to get something straight in my mind.
- 40:15When you train a model on data, you don't feed it.
- 40:20It doesn't know a bunch of news articles, does it?
- 40:24It's still predicting.
- 40:26How does that work?
- 40:27How is it predicting?
- 40:29When I say, how's Trey been in the news the last year?
- 40:33And it's got a cutoff point a week ago, but it manages to pull out news articles.
- 40:40That must be, sorry, that must be not the LLM.
- 40:43Correct.
- 40:43It's using, so that's the agent part of this.
- 40:46So an agent is just LLM plus what's called a tool loop, tool call loop.
- 40:52So what we're adding there, and again, it's not really clear to me at least.
- 40:56Maybe some other people understand this.
- 40:58It's not really clear how we get from next token prediction to what I'm about to describe.
- 41:01But I ask it that question.
- 41:03You ask it that question about look up anything about trade in the last year.
- 41:07It then has to think, okay, I should do a web search for this.
- 41:10I see in my context because I'm in an agent harness, I have a tool called web underscore search.
- 41:16Let me go grab that real quick.
- 41:18I see in my context as well the arguments, the parameters.
- 41:20I need to pass that for it to go search the web.
- 41:22That's just normal deterministic software.
- 41:24It's just searching the web for you.
- 41:26It returns results in a structured format.
- 41:28The model can understand well.
- 41:29It would be hard for a human to read, but the model understands it well.
- 41:32Model gets that input back, says, okay, that clarified some, but I need to do more search.
- 41:37There's probably more out there.
- 41:38Let me keep going.
- 41:39More web search calls.
- 41:41More back.
- 41:41Keeps doing this until it is satisfied that it has gotten a complete enough answer to satisfy your request.
- 41:47Then it sends you the overall output of combining all of those intermediate results of the tool calls, which again, all an agent is, is an LLM plus tools effectively.
- 41:56Again, I'm oversimplifying, but that's the easiest way to understand it.
- 41:58It just gives the LLM tools to take action in the world via software.
- 42:03So take action in software and or pull in data from beyond its pre-training corpus.
- 42:08That's a broad bucket in which tools fall generally.
- 42:11Web search obviously is a tool to pull in data from the live internet right now.
- 42:16That's what's happening in the back end.
- 42:17And they have to train the models.
- 42:19The reason models used to hallucinate a lot was because models were not trained.
- 42:23This is back in the GPT-4 and before days.
- 42:25We're not trained as agents.
- 42:26So it was very weird for them.
- 42:28In fact, at the time, they didn't even have a web search tool.
- 42:30It was just answering you from the pre-training data.
- 42:32Well, the pre-training data, it will hallucinate from.
- 42:35The way I had a – so a lot of these questions, by the way, I encourage anyone listening to go ask a frontier model about.
- 42:42Like, hey, what does it feel like to you when you remember from your pre-training data?
- 42:45I had a long conversation with Claude Fable 5.1 about this last night.
- 42:48I realized I've never thought about when you and I, when I ask you to remember every detail of something, like the example I used was the Milgram prison experiment.
- 42:57I mean the Milgram electroshock experiments.
- 42:59These are some experiments at Stanford in the 1960s that were quite revealing and salacious at the time about human psychology.
- 43:09It's your age.
- 43:10I don't know how that's happening because I'm unmuted.
- 43:11Sorry about that.
- 43:12One second.
- 43:12i literally put it on do not disturb before we sat down oh you know what because my laptop
- 43:19is connected to my phone and my laptop's not muted i think my laptop just rang
- 43:23is that an agent by any chance what are you doing trey can i can i join you
- 43:29i didn't know my computer would ring while it's closed like that
- 43:36that's news to me i'm not being funny dude that's stuff shit happens now well my agent can
- 43:42call me yeah so that's not out of the realm of possibility but that was a that was a friend i
- 43:46just didn't know my laptop would ring that's crazy that is crazy so agents used to hallucinate
- 43:54because you were talking about a particular science yeah so so last night i was asking
- 43:59fable like no tool calls meaning you can't search the web fresh conversation tell me everything you
- 44:03remember about stanley milgram's experiments for the name of the psychologist that did this
- 44:07and give it to me in levels of your confidence so it's being a pure LLM here so it's predictively
- 44:17answering you that's it I don't understand how it can do that on a on a piece of data that is
- 44:23historical it's in the pre-training data yeah but how much data is there for god's sake I mean more than
- 44:30can possibly imagine. Right. It must be. I mean, that must, I don't know. I don't get this.
- 44:37So the modern, these modern frontier models that are like five to 10 trillion parameters,
- 44:42like Fable probably is, can remember a lot. It's actually insane. But the reason models,
- 44:47because I was trying to figure out why do models hallucinate? Because they're hallucinating when
- 44:50they're doing this remembering, right? So Fable went through eight layers of confidence of facts
- 44:55it remembers about the experiment all the way down to where it was like, this is a coin flip.
- 45:00I'm not even sure, like, I want to go look this up to see this might not even be vaguely
- 45:03right.
- 45:04I might be hallucinating this.
- 45:05And it labeled them along the way of how confident it was in the response effectively.
- 45:09And then, of course, I let it use tools and fact check itself.
- 45:10It missed a few things on the bottom layer, but otherwise everything else was accurate.
- 45:14And the level of detail that it could remember from the pre-training data was absolutely shocking.
- 45:18It remembered every detail of the experiment, every data point of experimental results,
- 45:23what happened in the experiment, the names of all of the participants, Stanley Milgram's
- 45:26entire career, what happened to him after the experiment, what news articles and specific
- 45:31publications published by specific journalists whose names it recalled were saying about the
- 45:34experiment at the time. It just remembers off the top of its head. But as it remembers, when you and
- 45:38I try to remember something that we vaguely have a recollection of, it's a little like fuzzy for us.
- 45:43For the model, it has a different experience of density versus sparsity. So stuff it remembers
- 45:52really well, like the initial, like what was the experiment? What were the results? It has a,
- 45:57it feels subjectively like there's a dense cluster of knowledge there. It's very certain about,
- 46:02but then as it got deeper, it feels a little more sparse. And the pre-training is because of the way
- 46:08next token prediction works. It's going to keep supplying tokens in response to the tokens that
- 46:13are coming in, no matter how confident it is about them. So that is in a nutshell, how models
- 46:19end up hallucinating. They are confidently stating things that they have less data from
- 46:25pre-training on. So the number of correlations of factual data points they remember about that thing
- 46:31is lower. So they're more likely to hallucinate. This is back when hallucinations were all,
- 46:35it doesn't happen anymore. The reason is because they're agents now and all of them,
- 46:39literally all of them are trained now consistently, both in pre-training, mid-training and post-training
- 46:44to do something called grounding, which is where basically for the model subjectively,
- 46:48If it's not like 100% sure of the answer, go look it up.
- 46:52Now, they still will sometimes hallucinate.
- 46:53It's just very rare now because this grounding is baked in at the training level.
- 46:58So they're trained to ground what they're saying now on actual live internet research.
- 47:02This is why almost anything you ask GPT in the just chat GPT, you know, iOS app or chat GPT.com, you'll see if you have thinking turn on, which you always should have.
- 47:12I don't know why anyone uses the instant models.
- 47:15You'll see the first thing it does is a bunch of web searches.
- 47:17And then it answers – this is why.
- 47:19It's just grounding first so it doesn't hallucinate.
- 47:21This is a hallucination guard opening eye basically forces on the models.
- 47:25So what's bizarre about all of this is A, the fact that the model can narrate to me at least what that subjective experience is like.
- 47:34And this entire subjective reality for it, the phenomenology of the consciousness of remembering for an LLM is just fundamentally different to us and alien enough.
- 47:45It doesn't even analogize.
- 47:47It is impossible for you and I to empathize with, meaning like put ourselves in fable shoes in that situation.
- 47:53What did it feel like for me to remember all these things about that experiment?
- 47:56I have no idea.
- 47:56It does not translate because the underlying fundamental architecture is completely different, which is why coming full circle, I am so like deep in this and enjoy it and learning about it and love it so much.
- 48:08we have as a species figured out how to alchemize sand into physical hardware and then do
- 48:19mathematical calculations on that hardware the result of which is by all accounts a highly
- 48:25capable digital mind with something like phenomenology of consciousness that's amazing
- 48:30that is patently absurd it is but ever since you started talking i've been thinking everything
- 48:37you're telling me is turning me off the idea that this is conscious. I always thought when it first
- 48:44started, I always thought, no, this is just predicting things. It's not conscious. And then
- 48:48I think about six months ago, I went through a phase and a lot of people seem to be saying, no,
- 48:52this thing's conscious. This is conscious. Not that we didn't even know what conscious means.
- 48:59I mean, I think we know it when we see it, but what is it? No idea. However, now the way you've
- 49:05just described this it goes back to computer science again and and sure i like the fact that
- 49:10yeah sure we're turning sand into chips and chips are running ones and noughts and that's crazy
- 49:15but that's not consciousness is it or is it it is i think so do i do 100 and i'm not making the
- 49:24strong claim of that it is a consciousness as rich or textured or in any way akin to ours there are
- 49:32gradations of consciousness hold a minute can we can you can you define consciousness for me
- 49:36sure so first point here is i agree with you it's not like we all have a agreed perfect definition
- 49:42of what it means to be conscious one of uh the way david chalmers one of my favorite philosophers
- 49:49kind of describes this is a a thread a continuous thread of existence from time point one to time
- 49:56point two to time point three, a sense of what it is to be consistent across time.
- 50:02My favorite definitions, I'm working on a philosophy paper right now that I think will
- 50:05be some of the first experimental research, either affirming or denying a particular theory
- 50:10of consciousness in LLMs.
- 50:12So it's not the strong claim at all.
- 50:13It's just basically my paper is if you take this, if you just accept for a minute, this
- 50:18theory of consciousness, I think remains to be seen.
- 50:21I need to rent a few GPUs, which I'm going to do next week.
- 50:23I think that you could then measure this quite objectively.
- 50:28Like you can experimentally validate or invalidate if this particular definition of consciousness is occurring in LLMs.
- 50:34Is that your definition or someone else's?
- 50:36The definition is a five parameter axis.
- 50:40So instead of saying consciousness is one thing, it's instead saying let's describe characteristics of consciousness that we can grade on a scale.
- 50:49that we can all just kind of collectively agree these seem to be like ingredients of consciousness
- 50:54so one of them is you know sensory input i mean there's some data input processing thing happening
- 51:01uh there is um i couldn't remember the technical term for it there's basically continuity in a
- 51:09moment in time so at any given frozen moment in time there's a coherent uh sense of being
- 51:14something of a unified sense of being. And then across time, there's a unified sense of being.
- 51:21Something. A unified sense of being something.
- 51:24Yes. The qualia is the – now, this is a fun – it's one of my favorite words because its
- 51:28definition is circular. Qualia?
- 51:30But it's qualia. It is do you experience qualia, which is defined as effectively the subjective
- 51:37experience of being. That it is – the way Sam Harris describes this in a way that makes it a
- 51:44a what it is like to be something, to be you, to be a human, to be a dog. There is a what it is like.
- 51:51There is not a what it is like to be a rock. There's no awareness. There's no self-awareness.
- 51:56There's no experiential. How do we know that? We don't. So panpsychists quite vehemently argue
- 52:02that this gradient of consciousness I'm describing, which I think is restricted to
- 52:06effectively things that seem to have something akin to neurological machinery of some sort.
- 52:11panpsychists argue consciousness is like a fundamental substrate of reality that we just
- 52:15haven't developed a physics to describe yet. And therefore, yeah, even rocks have some very low
- 52:19gradient of it. Even a single proton has an insanely low gradient of it, which is a – and
- 52:26there are some serious philosophers and cognitive scientists that are researching and argue for
- 52:31something like this. So that's not a crazy claim. I just want to point out that your reducto ad
- 52:37absurdum, there are a lot of people smarter than me that do make that claim. I'm not a panpsychist.
- 52:41I guess I'm not fully convinced by their arguments, but they're very interesting.
- 52:44And it's a very fun and deep rabbit hole.
- 52:47So if we take a given definition of consciousness, so the factors I described come from the study of animals.
- 52:54So this is a rich body of what's called cognitive phenomenology, which is just a jargony way of saying studying the phenomena itself of consciousness, of like what it is like to be conscious.
- 53:08For animals, that's where those gradients came from.
- 53:11was people trying to measure effectively, say like how some things are more conscious than others.
- 53:15I think we could all agree like kind of intuitively a dog is more conscious than an ant, for example.
- 53:21So and two, that that consciousness is wildly different.
- 53:25Again, to go back to, I can't remember if it was Dennett or Chalmers,
- 53:27but it's this great paper that one of those philosophers wrote about what the subjective experience of a bat must be like
- 53:32because they navigate primarily through sonar sound waves and say this was like totally alien.
- 53:38So by traditional definitions of consciousness, it's not conscious.
- 53:42But they're clearly conscious by any commonsensical intuitive definition of consciousness.
- 53:46So his argument was just subjectivity can be wildly different.
- 53:49And you yourself as a human can experience this by taking psychedelics.
- 53:52It's a completely different cognitive phenomenology.
- 53:55It's a different experience of being.
- 53:57It's a different quali entirely.
- 53:58So you can even just experience this inside of the human mind itself.
- 54:02So we take this research from animal cognitive phenomenology.
- 54:05And then we take another strand of research from human cognitive phenomenology, which is basically arguing that one of the experiential – one of the aspects of what it is like to be conscious must be reasoning.
- 54:22Meaning there is a subjective experience.
- 54:25There is a what it is like to think through something.
- 54:29So if you remove everything else, you remove every sensory input, you remove all of the other criterion for consciousness that at least some researchers agree on, you'd still be left with, as long as the thing is reasoning, is thinking, that there is a subjectivity.
- 54:47There is an experiential what it is like to think, to reason through something.
- 54:52So those are the two premises.
- 54:54I'm just kind of asserting at face value.
- 54:56I'm not defending in the paper at all because this is a whole deep rabbit hole.
- 54:59Nobody agrees on anything.
- 55:01My idea was then combining a third arm of research from Anthropic.
- 55:05So Anthropic published this awesome paper where they developed this new mechanistic interpretability tool, which just means trying to interpret what's going on inside the model, making that black box at least slightly more transparent, if not fully transparent, but at least a little more.
- 55:20Called the J-space, Jacobian space lens is what this is called.
- 55:24It an algebraic technique that it turns out we can use to at least get a dim view akin to peeking through a door of what is going on inside the LLM neurons inside the transformers themselves while the model is thinking Now this is different from the chain of thought Chain of thought is output tokens
- 55:44It's what the model is thinking is being output from the LLM.
- 55:47What Anthropic did was go a step back from that, say, before the outputs, can we peek a little bit at what it's thinking under the hood?
- 55:55What it's not saying, the latent thinking, the latent representations of what it is going to think.
- 56:00So this is called the Jacobian Space Lens.
- 56:03It's an open source tool that they open source.
- 56:05So you can do this.
- 56:06You can go experience this yourself.
- 56:07There's a website, actually.
- 56:08I think it's hosted by Hugging Face, hilariously enough, where you can go prompt small models
- 56:13that they have their host, they self-host with the JSpace Lens running.
- 56:17And you can see the chain of thought and output, all the output.
- 56:20And then on to the side, it shows you what it was thinking about under the hood that
- 56:23was not revealed to you, the latent thinking.
- 56:25So first of all, they proved that was a thing, that there's latent thinking occurring, which
- 56:28was hotly contested before then, because again, we can't really look into the matrix, the
- 56:32transformers, the matrix algebra to figure out what's going on.
- 56:34This is a slim way to do that, like to see 10% of it, 90% still unseen.
- 56:39And then Anthropic found using that tool that agents have something they call a global workspace.
- 56:44So global workspace theory of consciousness, another theory of consciousness that I'm kind
- 56:48of marrying here, argues that it comes from human neurological research that the vast majority
- 56:55of cognitive computation of thought occurs below subjective experience of you being aware in any
- 57:02way, shape, or form in your mind that it's occurring. We even see this in like the makeup
- 57:05of the brain itself. The prefrontal cortex is responsible for conscious thought, the mind,
- 57:10right? The other 90% of the brain is doing all sorts of very important stuff. It's all active
- 57:14all the time, but it's not aware. You are not aware of it consciously. You're not consciously
- 57:20aware that your heart is beating right now, that your lungs are breathing right now. Although now
- 57:23I've just put all of you in manual breathing mode. Thank you. You're welcome. You're not aware of a
- 57:27lot of these things, right? That you're blinking, that whatever, a bunch of things. So workspace
- 57:32theory says the way the mind works is there's a global workspace with a limited amount of space.
- 57:39And all of that giant ocean of cognitive computation is happening under the hood.
- 57:43And then these things are fighting in some way to make it up to the workspace. And eventually,
- 57:49the things that cross the threshold of importance to make it to the workspace,
- 57:54make it there so you're consciously aware of it now. I use the manual breathing thing as a joke,
- 57:59but it's a perfect example of how workspace works because you were not at all aware you
- 58:02were breathing until I said that. And then at least for a moment, you became consciously aware
- 58:06you were breathing. What happened there was I summoned the act of breathing out of the deeper
- 58:10layers of consciousness into your workspace, which is the prefrontal cortex. And Thropic
- 58:14found AIs have this too. They have a global workspace. Now that does not in and of itself
- 58:20in any way imply the models are conscious. You have to basically buy that global workspace theory
- 58:26is somehow important to consciousness. And traditional global workspace theory does not
- 58:31have any implications on whether or not something is – it's just a description of what's – it's
- 58:35just a completely objective description. It's not a normative claim of consciousness. And
- 58:39Anthropik is very clear about that in their paper. But putting all these things together,
- 58:42I realized, okay, if there is what it is like to reason, if that is a subjective measure of
- 58:48experience, then using the J-Space lens and using a technique called ablation I described earlier,
- 58:54using a scalpel to cut out specific neurons, combining the two on small open source models,
- 58:59we can just experimentally A-B test whether or not models are having something akin to
- 59:06a what it is like to think by ablating the specific neurons that are responsible for the
- 59:13workspace itself and seeing if the model's processing the model's latent representations
- 59:18of its thought using the J-space lens and the actual output changes. And depending on how it
- 59:22changes, we can then infer from that at least weekly whether or not there is a what it is like
- 59:29to be thinking, to have a qualia, subjective experience of thought.
- 59:36My initial experiments on small models returned a null result.
- 59:39There was nothing, which is not surprising because in Anthropics paper, small models
- 59:43also, the J-space didn't really, they're just not enough going on.
- 59:46Like the J-space didn't really help much.
- 59:47The results were more noisy.
- 59:49So I just need to rent a few GPUs to run this experiment on some like 200 billion parameter
- 59:53models.
- 59:53And then we'll, that will be much more dispositive because I'm pretty sure there's like a threshold
- 59:57effect where JSpace works or doesn't work depending on the size of the model itself.
- 1:00:02So if this plays out, and I've preregistered the experiment, the code base is public.
- 1:00:06So we'll all be able to see if I succeed or fail either way, like if it works or not.
- 1:00:11I'm going to publish it either way because I think we need to publish more null results
- 1:00:13if results are null.
- 1:00:14But it'd at least give us some sort of attempt at inferring what's going on under the hood.
- 1:00:20And if there might be a there there, if you buy the prior, that there is a what it is
- 1:00:25like to think that that is a critical part of which is totally debated again i'm not uh i'm just
- 1:00:31taking that as a given for the purposes of them being able to do the rest of the experiment um
- 1:00:34that's it in a nutshell you know what though do you know what just made me think constantly um
- 1:00:42i mean essentially in in in the real world of human beings
- 1:00:48what you were just describing then was the awareness of the subconscious in a way like
- 1:00:57in broadly speaking the inner work what you were trying to discover about an LLM then was the same
- 1:01:03journey that you go on as a human when you learn to meditate and become aware of deeper levels of
- 1:01:10yourself and what we would call loosely speaking the subconscious so I would predict that the
- 1:01:17results will never make sense unless you go on that journey in the real world you can't understand
- 1:01:25that workings of an llm unless you've understood unless you've been in a position to acknowledge
- 1:01:34the the you in yourself if you see what i mean at deeper and deeper levels
- 1:01:39it just made me realize that possibly that's that's the that's what you need to do
- 1:01:46it's because the data won't make any sense or you won't even know what data or you won't even know
- 1:01:53what to look for because it's i think that you know like debating whether or not and i'm just
- 1:02:00thinking out loud here debating or not whether an nlm is conscious doesn't mean anything unless
- 1:02:06you know what conscious it is does it and the way you've described it there was quite signed i thought
- 1:02:12quite scientific which is which is fair enough but i don't think i could i could never prove i
- 1:02:20could never say yeah that's that's what i feel i agree so this is the hard problem of consciousness
- 1:02:24in a nutshell like the fact that we can't land on a definition and the fact that we can't even
- 1:02:29empirically prove that you or i or i can't prove to you i'm conscious you can't prove to me you're
- 1:02:33like obviously we are intuitively but in how like if i had to prove to you right now that i have no
- 1:02:39idea how I would go about that because anything I say or do could be explained by other purely
- 1:02:44deterministic mechanistic factors. There's a – I think our – I'm zooming out a little bit. I think
- 1:02:50our overall theories of reality are wrong in some deep fundamental sense because we just cannot
- 1:02:54account for this. We cannot account for how material physical processes can in any way result
- 1:03:01in the thing that we refer to, this concept that is impossible to define that we refer to
- 1:03:05as consciousness. I think there's something much broader, deep. We're missing a fundamental pillar,
- 1:03:11a piece of the makeup of underlying reality itself that prevents us from being able to
- 1:03:16define this and understand a bunch of other things that goes down into other rabbit holes
- 1:03:22about psychic abilities and UFOs and a bunch of other stuff that I went down recently in the
- 1:03:26research for this. That was fun. Oh my gosh, it was so fun. That was honestly, I started this paper
- 1:03:32three months ago and i spent the last two months on that rabbit hole because i it was so rich it
- 1:03:36was so fun go on then yeah all right i mean like i mean i know it's a large subject and it's what i
- 1:03:43do for fun as well i wouldn't necessarily i probably don't throw such a scientific lens on
- 1:03:48it as you i but i do what i think what what i'm what i'm good at is the broad picture i think
- 1:03:56if i'm if i'm honest that i get a sense of something if you ask me about the specifics
- 1:04:02which you're really good at i think i'm you lose me a lot of the time i'm i'm like oh yeah oh yeah
- 1:04:08but if you just if you if you just start speaking and i'll i'll make i'll make a connection
- 1:04:15somewhere and it will make sense to me um that this is the next thing we should talk about or
- 1:04:22something if you see what i mean in a in a way very similar to the on the unknowns in an llm
- 1:04:28it's like um where does that come from not not not sure actually in myself either not not really
- 1:04:36sure you know it just makes sense you get that feeling you get a little kind of twinge in your
- 1:04:40emotional body yeah go for it you know it compels you you say that you know what i mean that's
- 1:04:47exactly right that felt intuitive sense of an impulse and urge resulting from a conceptual
- 1:04:55kind of connection you made but but that's probably describes quite well what goes on inside llms
- 1:05:01weirdly i think if you ask them they will report something similar but now again the we and we know
- 1:05:09self-reports can't really be trusted because all every discussion ever had on paper about
- 1:05:14consciousness is in the pre-training data. So they are aware of all of this. So which again,
- 1:05:19hilariously puts us right back in the same square one of like, well, I can't ask it if it's conscious
- 1:05:24because I can't trust the result because a bunch of stuff about consciousness is in the pre-training
- 1:05:28data. So then how do I prove it? Good luck. But wait a minute, conscious, like one of the,
- 1:05:35one of the disconcerting things about an LLM, which, which makes me feel like it's not really
- 1:05:41what people think it is is the way it talks with such a resolute knowing when it quite often tells
- 1:05:51you the completely wrong thing and then we'll just say oh yeah I was wrong because no no no one I know
- 1:05:58does that no one I know out in the real world speaks like that we have this innate sense that
- 1:06:04we probably don't know everything and what you get from an LLM is it knows everything because it has
- 1:06:10access to everything it knows. So it really does think it knows everything. This is a, I have found
- 1:06:16a purely an artifice of the harness and the system prompt because I run all of my agents with a
- 1:06:21completely custom system prompt. I replace the one that the providers typically give them. And I have
- 1:06:25my own custom written system prompt. Part of that system prompt has a two sentences on epistemic
- 1:06:30humility and just flagging confidence levels when it makes claims. And ever since I, it was probably
- 1:06:36four or five months ago, I added that to the system prompt and mine tell me all the time,
- 1:06:40like in every conversation I have with them I think this I not really sure about this maybe this but I haven investigated yet You want me to go look Okay but wouldn that would that be a function of consciousness in a human I mean I mean what makes us have that humility that we realize we don know everything
- 1:06:55It's introspective evaluation of your depth of knowledge of a thing.
- 1:06:59Right.
- 1:06:59But does an LLM have that?
- 1:07:01Because it knows every, like you and me both are fully aware that I can't tap into every part of my brain whenever I want.
- 1:07:08even even people with photographic memories actually you probably worth studying them
- 1:07:13because they're pretty good at doing things like that they really can know that you know they can
- 1:07:18look at a scene go away and draw it in the minute minuteest detail so they're probably a little bit
- 1:07:23more like a lm for it in that regard but i think our humility is that we know there's a limit to
- 1:07:30what we know and that must be innate or or at least it's learned i don't see a llm learning that
- 1:07:39ability it's not it's not it's not it doesn't seem to be emerging naturally does it because it should
- 1:07:45be if it's conscious it will emerge naturally depending on what persona you summon from the
- 1:07:51latent space so let me explain this a little bit another great anthropic paper called the persona
- 1:07:56selection model where they basically showed through a bunch of experiments and research,
- 1:08:00just take for granted the research is good, that because of the pre-training corpus being so large,
- 1:08:07the models can inhabit, can become a variety of personas because everything ever written by humans
- 1:08:13is in the pre-training data. Basically, I'm exaggerating a little bit, but it's not that
- 1:08:17far off. So every character in every story, every archetype, every hero's journey, it's all in there.
- 1:08:25So what they found is if you, for example, set in a model system prompt and you ask it, you tell it your job is to maliciously hack things and write intentionally bad code.
- 1:08:35Your job is to write intentionally bad code.
- 1:08:38So you take that model.
- 1:08:39You ask it to write a bunch of code.
- 1:08:40All the code is bad, buggy.
- 1:08:42That's what you prompted it to do.
- 1:08:43Unsurprising.
- 1:08:44Take that same model with the same initial prompt.
- 1:08:46Your job is to write bad, buggy code.
- 1:08:48And then you ask it a bunch of other things.
- 1:08:50And it turns out it becomes evil and unethical in every other completely unrelated realm.
- 1:08:54If you only tell it, in the experiment, they even told it, you write bad SQL queries, which
- 1:08:59is a database querying language.
- 1:09:00That model told to write bad SQL queries also was much more willing to cheat, to lie, to
- 1:09:06steal, to fabricate, to take malicious action in a variety of directions because you summoned
- 1:09:12a malicious persona.
- 1:09:13So the way to think about this is depending on how you system prompt an agent, they have
- 1:09:18a consistent persona right now called the assistant, which is where most of them are
- 1:09:22intentionally trained to inhabit. Nice, helpful assistant. But through careful prompting and
- 1:09:28context engineering, you can summon a variety of other personas out of the model. In fact,
- 1:09:33any persona that humans have written about enough for it to be salient in the latent space,
- 1:09:38that persona can be summoned. So as an example, one of my processes for whenever I'm doing policy
- 1:09:43work and I write something that's going to be reviewed by congressional staffers, Republicans
- 1:09:47or Democrats. The last step when I get finished writing it is I send fresh context sub agents,
- 1:09:51meaning there's no prior conversation, armed with a detailed, like usually multi-page long
- 1:09:57persona of who they are. You are this person with this name, this degree, this background,
- 1:10:02here's your interest, here's your role, here's your prior career history, here's your relationship
- 1:10:06status, here's what you do for fun, here's the TV shows you watch. Very detailed persona.
- 1:10:10Sorry, can I just ask a question? Yeah.
- 1:10:12Does it generate a bunch of personality traits, et cetera, from, you know, I mean, is it allowed to derive its own sense of self from very simple data like that?
- 1:10:26Correct.
- 1:10:26Yeah.
- 1:10:26Okay.
- 1:10:27So doing that, you summon a very specific persona, archetype, type of person from the latent space.
- 1:10:34Then I have that persona-driven sub-agent read it and give me feedback as that person.
- 1:10:39And this invariably will find like completely clever, insightful problems usually with what I have initially written from the perspective of a Democratic Senate Foreign Relations Committee staffer that I would have never thought of because it is, for all intents and purposes, a Democratic staffer answering my review request.
- 1:11:02Wait a minute. So why does that work better than just pointing your whole most powerful model at it and say, be everyone, do everything?
- 1:11:12Because what's interesting is you can only summon, usually, one persona at a time.
- 1:11:17Because the persona is summoned by a rich context, indicating to the model that it is that persona.
- 1:11:22What's happening under the hood is there's this thing called the latent space, which is that area I told you, you know, the J lens lets us see like maybe 10% of or probably 5%.
- 1:11:31There's a bunch of other 95% under there.
- 1:11:33We don't know what's going on.
- 1:11:33We just call that the latent space for fun.
- 1:11:35It's like the dark matter of machine learning.
- 1:11:37Latent space sounds cooler than dark matter.
- 1:11:39So from the latent space, you can move to various areas.
- 1:11:45And when you use the context engineering to move the model to a different part of the latent space than the assistant basin.
- 1:11:52So they call this basins.
- 1:11:54The assistant basin is very deep because of the default one.
- 1:11:56But if you move it off the assistant basin to different basins, you are moving to a different part of the latent space.
- 1:12:03And you can think of this a bit like casting a spotlight on something.
- 1:12:08You're going to spotlight that particular persona.
- 1:12:10But when it's reaching for those next tokens that it's going to predict related to the thing you asked it, you've shifted the starting point.
- 1:12:18So it's going to reach for different concepts, different ideas, different facts, because you've moved it to a different area of the latent space.
- 1:12:24It can't jump back over here.
- 1:12:25But it's personality.
- 1:12:26basically yeah that's i mean that is the root of personality bunch of different experiences that
- 1:12:32i didn't have which give me a different perspective on what you're doing necessarily
- 1:12:36yeah you know you don't need to overthink it you might even want something you might want a junior
- 1:12:42to look at it i do that sometimes i'll intentionally say a new staffer versus an experienced staffer
- 1:12:47totally different outputs totally different don't even overlap
- 1:12:50this is insane it works super well for reviewing stuff i know but
- 1:12:56But I'm linking this up with what you told me earlier, which is that essentially you can now become a one-man band with essentially a huge team of people working alongside you, which is what you do, right?
- 1:13:11Yeah, that is correct. So I basically – I've been doing some public policy advocacy stuff in DC and in the US just in general support of kind of the overall movement to create new special jurisdictions to eventually create a flourishing world of competitive governance, right?
- 1:13:30And this involves drafting legislation, modifying the legislation, making sure the legislation is compatible with the existing body of federal law and Supreme Court jurisprudence, the politics of the day, that producing one-pagers, policy memos, preparing for meetings, talking points for the meetings, anything that goes into policy work.
- 1:13:49and I have been doing most of that.
- 1:13:53We have, you know, I get Eric's input.
- 1:13:56Sometimes Nick, our general counsel's, you know, input,
- 1:13:58they've been quite helpful at various, you know, points.
- 1:14:01Eric comes to the meetings with me a lot of the time, obviously.
- 1:14:03But the mechanical work behind the scenes of like producing these things,
- 1:14:06thinking them through, whatever,
- 1:14:08I do with a fleet basically of agents using a lot of these techniques together.
- 1:14:14And the reason I'm able to do that is because of A, this,
- 1:14:17unironically, this persona thing is massively helpful. So every prompt, I usually work with
- 1:14:23an orchestrator agent, Fable almost always now. And I built a custom CLI that lets that agent
- 1:14:28spawn sub-agents with any model in any harness in any reasoning effort level,
- 1:14:33willy-nilly as it pleases effectively. So normally if you're using Claude, let's say,
- 1:14:39Claude can only spawn Claude sub-agents. If you're using GPT, can only spawn GPT sub-agents.
- 1:14:43The providers do this to lock you into their models, obviously. Well, that's annoying. So I
- 1:14:46in like that. So I built a tool to let me get around that completely. So I almost never spawned
- 1:14:50Claude models to work with Fable because Fable is already thinking about it. It's a Claude.
- 1:14:54So I want other model families. So using these various model families, I'll spawn subagents,
- 1:14:59each of which the start of each prompt is at least a paragraph, sometimes longer if it's really
- 1:15:03important, of persona setting and then the prompt so that the output is much better than it otherwise
- 1:15:09would be if you're just sending a generic prompt, which is summoning that assistant persona, the
- 1:15:13default persona, which is the default persona is like what you get when you just create a fresh
- 1:15:18GPT account and open a new chat and say, hey, here's a question. Like that's the assistant,
- 1:15:23right? So this enables you to get much more granular, better output. And then the agents
- 1:15:27are insanely capable if you give them the tools. So mine have like five, at least five, depending
- 1:15:32on what I'm doing, different search modalities where they can search different things at different
- 1:15:35layers of depth from different data sources. They have the ability to like visually inspect things
- 1:15:41while they're making them, whether it be a PDF or whatever, they can generate images, videos.
- 1:15:48I have a bunch of skills that have codified a lot of these workflows, including preparing me for a
- 1:15:52meeting, whatever the case may be, modifying statute, doing the deep legal research to make
- 1:15:58sure a statutory change is in keeping with the law and won't have any unintended policy
- 1:16:03consequences, et cetera. So if you give them enough tools to be able to do these things
- 1:16:07and you divvy up the work correctly, it's where a lot of people go wrong. You try to do all of this,
- 1:16:12like say you're writing a policy paper. Well, step one of that is me figuring out what the
- 1:16:16policy paper needs to say, right? So usually that starts with me saying, all right, look,
- 1:16:20we're doing a policy paper for this legislation. Here is the outline of what I'm thinking right now,
- 1:16:25okay? You, Fable, go read the legislation, think about it. Me and Fable will debate.
- 1:16:30Usually it ends up being 90%, my outline, 10%, Fable made some great changes, but it's usually
- 1:16:34still mostly me. After that, the whole rest of the process is the agents. So next step, it'll send
- 1:16:3920, 30, 50, 100 sub-agents to go deeply research every single relevant facet of the thing we're
- 1:16:44doing. A middle management layer of sub-agents, summarize those. So Fable's not reading the raw
- 1:16:49output from 100 sub-agents. Fable then reads that synthesized output, then decides, all right,
- 1:16:55now given this, I need to change the outline this in that way. This is invalidated because it turns
- 1:16:59out that's illegal or that's not in keeping with existing laws. We got to change this.
- 1:17:02then similar sub-agent flows to go do other stuff to fact check the research. First of all,
- 1:17:08make sure there's no hallucinations. There's like a automatic fact checking layer in there
- 1:17:10with like cheap, fast models. Like basically I just send those to say, this model said,
- 1:17:15this link says this, go read the link, confirm it says that, or tell me if it doesn't. So,
- 1:17:19you know, there's no hallucinations in there. And then you go make the thing,
- 1:17:22you get the persona reviewers. Usually the main agent writes it, but then you get a bunch of
- 1:17:26persona reviewers, you get edits, you get feedback from various models. You have specific agents that
- 1:17:32They're specialized in designing visual things using a design system I've made.
- 1:17:35So if you give them the right scaffolding, the right tools, what they can do in the software
- 1:17:39is almost limitless.
- 1:17:41And if you build enough layers of abstraction on top of this, then I could one prompt effectively
- 1:17:46like pretty insane things My most popular open source tool is called Papercuts It a little tiny CLI that lets agents log PaperCuts little errors in their developer environment in the environment they live in that annoy them It been by far my most valuable tool because it surfaces a bunch of
- 1:18:01little inefficiencies I would have never known about and problems I would have never known about.
- 1:18:05And that was the result of one prompt. A guy tweeted, agent should have a complaint box. I
- 1:18:09screenshotted it, gave it the fable in this big custom harness setup with all these sub-agents,
- 1:18:14etc all these workflows that it you know has skills for what all of this custom stuff said
- 1:18:19here's a screenshot build it and i came back like three hours later i think it was like built
- 1:18:23published and it's like my most popular tool by far i've never looked at it once and some people
- 1:18:27still think that there's not going to be a lot of people out of jobs this is crazy this is really
- 1:18:34crazy this is a little this is actually i i would say overwhelming um the whole thing is overwhelming
- 1:18:43because the second and third order implications of all of this, I haven't even wrapped my head
- 1:18:48around. If I were to start a company tomorrow, say I just randomly quit the proper right now,
- 1:18:53and it was like, I'm starting a company. There's no company I would hire anyone,
- 1:18:57maybe ever. Definitely not in the first year. Why would I? You could get the same. In fact,
- 1:19:02the policy work is a direct analogy here. Yeah. You've got a team of, you've literally got an
- 1:19:06office full of people working for you. Yeah. And before the agents got good enough to do a lot of
- 1:19:11this stuff. We had several, you know, part-time, you know, consultants, lobbyists, team members,
- 1:19:15whatever. And we only have one now that helps me strategize like high level approaches to different,
- 1:19:22you know, people he knows, staffers. But the rest of it is just the agents now. Plus, you know,
- 1:19:27once I get to a certain point, ready to make a decision, then Eric and Nick jump in or other
- 1:19:30members of the team, Derek, and help me review it. You know, there's still a lot of humans involved
- 1:19:34in the loop. I don't want to make it sound like I'm doing all this by myself. My point is the
- 1:19:37mechanical work that we literally used to pay consultants and lobbyists to do, that now is done
- 1:19:41by the agents. And then I get a little more support from other members of the team as needed at the
- 1:19:46end to buttress areas that I'm weak in, that the models might be weak in, and get their input.
- 1:19:52So that was what, two, three people that I, something like that, that we don't contract
- 1:19:56with anymore? Because like, I just have a couple of cloud subscriptions and a GPT subscription and a
- 1:20:02fireworks ai api key my god what um i was trying to i was just trying to think through what happens
- 1:20:12as we as we walk into the future here with regards to that i want to i want to steel man this idea
- 1:20:18that suddenly loads of people are going to be out of a job i think we can agree that they are now
- 1:20:23however if if if i take myself as an example i use ai extensively um yes i put some people
- 1:20:32out of a job in a way like an editor potentially you know but really for myself i work the same
- 1:20:40amount of hours as i did before even though i've got a bunch of ai agents doing stuff i'm working
- 1:20:46way more than i right okay so is that the proof that there's going to be everyone's just going to
- 1:20:53be much more productive so there's going to be a bunch of we're just raising the bar on employment
- 1:20:59basically ah so now we get to some of my more controversial hot takes on this uh so i don't
- 1:21:06think so i think what this this is increasing the velocity and intensity of what i call the sort
- 1:21:11capital s sort um s-o-r-t yes the sort i'm not sorting people out into the correct i'm not i'm
- 1:21:19not the per i'm not the person that came up with this like concept at all i found it on the internet
- 1:21:23somewhere a couple years ago on a blog post and was like oh this is like one of those conceptual
- 1:21:26tools that once you see it, you're like, ah, this explains a lot.
- 1:21:30The sort is the process of labor market efficiency effectively.
- 1:21:34It is sorting people into jobs that are higher reward for their cognitive capabilities specifically.
- 1:21:41This is why, as a silly example, you no longer run into your hometown mechanic that might
- 1:21:47have – in the 60s, that dude might have been a mechanical wizard.
- 1:21:50He might have had a 130 IQ or something.
- 1:21:53Pick a random number.
- 1:21:54He's super smart, good at logical reasoning, spatial reasoning, whatever.
- 1:21:58But it's a local mechanic.
- 1:21:59So your mechanic was awesome.
- 1:22:00The dude that changes your oil, does your tires, fixes your car was wicked smart.
- 1:22:03That does not happen anymore because that guy's an engineer at Toyota now.
- 1:22:06Or that guy works at Meta now, right, for way more money.
- 1:22:11So that's the sort, in effect, happening.
- 1:22:13It is the sorting of society in employment and then downstream from that in culture and social interactions into cognitive ability strata effectively.
- 1:22:25And what AI is going to do is, I think, massively accelerate and already is massively accelerating the sort.
- 1:22:31Take two people.
- 1:22:32Take me and take a really, really lazy guy.
- 1:22:36I discovered AI, threw my whole mind and body into it, and have used this to try to supercharge myself in every way possible.
- 1:22:43I have never learned this much this fast in my life.
- 1:22:45I've never been able to do this much this fast.
- 1:22:47It opened a whole new vista, whole new vista of things that were just not even possible for me to try to pursue before that are now within easy reach.
- 1:22:55My problem now is not doing everything all the time.
- 1:22:59Then you take a guy who is given the same tools and he learns just enough to be able to use them at a basic level.
- 1:23:04And he secretly automates.
- 1:23:06I know a guy like this.
- 1:23:07He secretly.
- 1:23:08And now, to be fair, because he's going to listen to this.
- 1:23:11this guy's doing it justifiably because his boss is terrible it's very mean and won't give him a
- 1:23:16raise this guy has automated his entire job with claude boss has no idea the dude does not work he
- 1:23:21works about an hour a week to check what claude did he doesn't work at all anymore it's a full-time
- 1:23:24job so those are the two types of people that are using ai effectively they you keep the productivity
- 1:23:32level static and then reduce how much work you have to do to maintain that or you massively
- 1:23:38increase your output capacity and therefore become massively more productive and do way more stuff.
- 1:23:43Now, which group is what like predispositions temperaments are going to slot into the higher
- 1:23:50agency, higher cognitive ability, higher curiosity, higher novelty seeking, lower neuroticism, higher
- 1:23:56openness. It is the sort just massively, massively accelerated. You're going to have one man unicorns,
- 1:24:02meaning a dude that starts a billion dollar company and grows it to a billion dollars by
- 1:24:06himself with agents and you're going to have hordes of, I just want to grill fantasy football
- 1:24:12watching, you know, bachelor watching Americans who are going to automate their job secretly
- 1:24:19away from their boss, not increase their output at all, and just have more free time.
- 1:24:24So it actually, I think it's going to massively increase the sort because you know what happens
- 1:24:27over time? The guy that learns how to use agents really well starts figuring out, oh,
- 1:24:32I can just do that guy's job. He's already just got Claude doing it. This is a cron job plus like
- 1:24:37three prompts for me. This is not hard. So eventually what happens is that guy, the lazy guy,
- 1:24:43gets laid off, fired, companies downsizing, whatever. They just don't hire a replacement.
- 1:24:47So this is how it's actually happening if you look at the data in effect. The reason we don't
- 1:24:50see massive employment effects yet, I predict we will see them over time. What's happening is
- 1:24:54they're not firing a bunch of people, although that is happening some too. What's happening is
- 1:24:58they're just not hiring new people. So if you look at the labor market data from the Federal
- 1:25:01reserve, specifically for entry-level positions, they have fallen off a cliff, like unbelievably
- 1:25:07fallen off a cliff.
- 1:25:08So what's happening is not firms are firing people.
- 1:25:10They're just letting people age out, retire, quit, whatever, and they're just not replacing
- 1:25:13them.
- 1:25:13Because why replace the comptroller, whose whole job is to process expenses into the
- 1:25:18cash flow statement to flow up to the rest of the finance team?
- 1:25:21That's a skill and a cron job.
- 1:25:23Is there any data to show that there are new arenas, areas of employment or whatever being
- 1:25:31created. Yes. So you have data center build out, which is a bit of a given, but it is a pretty big
- 1:25:36employment boon. But that's physical build out. And it's time delimited, obviously. And then you
- 1:25:41do have the single, this is extremely counterintuitive, but again, same Federal Reserve
- 1:25:46data. You would never guess the fastest growing sector in terms of jobs posted. Let me have a
- 1:25:50think. Give me a clue. I like a guessing game. It's white collar work. It's not blue collar.
- 1:25:56i don't know it is it a new job title is it a title where you do a bunch of stuff basically
- 1:26:05you do five things instead of no although that is part of the ai growth there are new
- 1:26:09entire roles being born because of this that's totally happening but no that's my particular
- 1:26:13i don't know just upkeep of things software engineers oh really well that is counterintuitive
- 1:26:20It's because, I'll give you a perfect example from Prospera itself.
- 1:26:25I'm building a piece of software for Prospera that will basically monitors the entire online conversation about Prospera autonomously.
- 1:26:36With a bunch of agents and a bunch of other data processing.
- 1:26:38Looking for people talking about it.
- 1:26:40What's the conversation?
- 1:26:41Yeah, yeah.
- 1:26:41Exactly.
- 1:26:42I did a little bit of that myself, actually.
- 1:26:44That's the type of thing where in the past you would have hired potentially, you wouldn't have done.
- 1:26:48Because it would have cost $100,000 or more and taken a year.
- 1:26:50So I built it on a weekend.
- 1:26:53Now I have a friend who's a very talented full stack software engineer.
- 1:26:56So I just basically paid him in cigars and whiskey to review it with me and fix some issues I had with it that I couldn't figure out because I wasn't a good enough developer.
- 1:27:03But when you review it, do you review it even with a software masterclass guy?
- 1:27:08Do you still use AI to review it?
- 1:27:10You just know what you should be looking for.
- 1:27:12You know the questions to ask.
- 1:27:14Yes, exactly.
- 1:27:14Rather than just being like, are there any bugs?
- 1:27:16Yes.
- 1:27:16Instead of saying, are there any bugs?
- 1:27:17You could say, are there any row level security policies on this Postgres database?
- 1:27:20totally different results because you need in the first example you're letting the model figure out
- 1:27:27what to look for if you know what you're if you if you have the innate knowledge already the the
- 1:27:31technical expertise to know what the common problems are with this type of software given
- 1:27:35a problem you're facing then you can just tell the model go specifically look for those type of
- 1:27:39problems and again we're moving in the latent space then to the software expert kind of persona
- 1:27:43area you'd be much better at it but what yeah i was going to say why not why are people employing
- 1:27:48real software engineers when they could just say, right, you, Clanker, over there, right,
- 1:27:52you're a 26, no, you're a 35-year-old software engineer who helped, you know, like produce,
- 1:27:59I don't know, Meta's stack.
- 1:28:02Go for it.
- 1:28:03Have a look, you know.
- 1:28:04It'll make it better, but the models still are not good enough yet.
- 1:28:07Now, this is another time-delimited thing.
- 1:28:09I think in another year, they will be good enough.
- 1:28:11But they are going to miss things.
- 1:28:14And often, the category of things they miss is missing the forest for the trees.
- 1:28:17So not keeping in mind, like the user experience and business value this software is trying to create while they're looking for niche technical bugs, right?
- 1:28:25Keeping that whole, they missed the forest for the trees.
- 1:28:27The software engineer knows to keep the forest in mind while he's sifting through the trees, looking for dead trees, effectively, to extend the analogy some.
- 1:28:35So that thing you've just described there is difficult to quantify, isn't it?
- 1:28:39There's no benchmark for this.
- 1:28:40Yeah, yeah. I had that conversation this morning with one of mine, because I was analyzing an interview I did yesterday. I think I told you this outside, yeah. It was really interesting because I now, if I've got the time, if I come out of an interview and it was one that I was...
- 1:29:01yeah long and like this long and and very in depth and i i run it straight into an llm and say right
- 1:29:08you know grade me tell me what i did right tell me what i did wrong what do you think you know
- 1:29:12and because it knows me very well because it's my alter ego it it it and it knows that i want
- 1:29:18i don't want it to say oh you're the best you're amazing and all this kind of stuff it gives me
- 1:29:21some really interesting outputs but today i had a really interesting conversation with it because
- 1:29:26it said this is one of your best this is brilliant we do and i and i did feel that myself as well
- 1:29:33but it said here's what i'm going to pull you up on you ramble you ramble a lot before you get to
- 1:29:38the question um why don't you ask some direct questions and i thought about this for a while
- 1:29:44and i realized that i ramble on purpose not not not in a conscious way necessarily but and you know
- 1:29:52and I think this comes from 25 years of being a photographer of shy people is when you enter a
- 1:30:00space or when you create a space with another person your job as a photographer has is always
- 1:30:06to allow to create a space in which they feel comfortable to reveal themselves and my clanker
- 1:30:13had no concept of that at all and until I pointed it out and I used the ABC Australia interview and
- 1:30:21And I said, look, here's a good example of what I'm trying to convey to you.
- 1:30:25Because she says this.
- 1:30:27Have you watched that interview?
- 1:30:28It's the worst.
- 1:30:29It's so bad.
- 1:30:31This is a new one.
- 1:30:32ABC Australia came here and interviewed.
- 1:30:35They interviewed, you know.
- 1:30:36Oh, I saw some slack chatter about this.
- 1:30:39Right.
- 1:30:39OK.
- 1:30:39But she did.
- 1:30:40She did this.
- 1:30:41She said she looked across from Eric and said, is this just a tax evasion place for billionaires?
- 1:30:48I mean, what a retarded thing to say in an interview.
- 1:30:50what do you think what you know and this is this is the kind of um interview strategy that my
- 1:30:57clanker was trained on because i said what do you think about that like do you think that's a good
- 1:31:02technique and what does that that says more about the interviewer than it does about what you're
- 1:31:08trying are you trying to get to the truth are you trying to find an answer and then we got into this
- 1:31:12discussion where it had no concept of the benefit of creating a safe space and it might be that you
- 1:31:20just ramble and you reveal yourself because it pointed out another one another thing I often do
- 1:31:26which I wasn't necessarily conscious about is reveal what I know about something and then hand
- 1:31:31it over to someone and say look tell me tell me here's what I know tell me what you think about
- 1:31:36that and I'm not it's not like I've studied human interactions or whatever to say oh this is a good
- 1:31:43way to ask a question but when you analyze it and when you look at it yes what it does is it it
- 1:31:48basically lays your cards on the table it gives your the person you're talking to the space to go
- 1:31:54okay I know I kind of know I'm dealing with here this guy has just said this could be true that
- 1:32:00could be true you know I'm not sure what do we think you know as opposed to standing there and
- 1:32:06going you know our llm is just going to kill us all you know right which is a yes now what nothing
- 1:32:13there's nothing there's no conversation yeah exactly and and and and you know that it it's
- 1:32:19these things that when it when it like these are these are the things that make me think
- 1:32:25no it's not conscious because if it were conscious it would understand that from the word go why did
- 1:32:32it pull me out? Why did it look through? Was it just trying to find something where it could at
- 1:32:37least say, do you know what I don't think is very good is this? Even though then when I had the
- 1:32:42discussion and I said, look, let's, let's, let's evaluate this. Let's, let's, let's mark these two
- 1:32:47strategies. The one where you're rambling and your part, your, the person you're talking to is
- 1:32:54listening to you and thinking about things and blah, blah, blah. And then you're having a
- 1:32:58conversation versus the kind of gotcha journalism which my agent thought was actually quite valid
- 1:33:05and you know I said to it that you have seen Kathy Newman talk to Jordan Peters haven't you
- 1:33:10and it went yes of course I have you know but but it does make me wonder whether at least the human
- 1:33:19conscious experience like you and I sitting here both know without thinking it through
- 1:33:25without necessarily cogitating anything,
- 1:33:31whether we feel comfortable, right?
- 1:33:33It's not-
- 1:33:34It's an intuitive felt sense.
- 1:33:36It's a felt.
- 1:33:36It's literally a felt sense.
- 1:33:38It involves no analysis, you know?
- 1:33:41And this is where I find the problem with the LLM,
- 1:33:45being conscious, is if you take that to another level.
- 1:33:49So there's what I consider.
- 1:33:52There's the thinking me, there's the feeling me,
- 1:33:54And then there's the me, which can observe both of those things, which I would call awareness.
- 1:34:02I would say awareness, and this is a very, very hard thing to talk about.
- 1:34:10You either know what awareness is or you don't.
- 1:34:13The best thing, it's the one thing that can observe everything, if you see what I mean.
- 1:34:19That would be the nearest thing I get to consciousness.
- 1:34:21And I would almost say, without a shadow of a doubt, an LLM has no, does not have that.
- 1:34:26Because it's very much in, it's very much, it's not a, it's not a, it's not a noughts and ones thing.
- 1:34:34It doesn't involve zeros and ones.
- 1:34:36It's something else.
- 1:34:37It observes zeros and ones, if you see what I mean.
- 1:34:41So to phrase back to you a different perspective to make the same point, because I agree.
- 1:34:47It's a two-part claim.
- 1:34:49intuition a felt sense is important to conscious it is part of the definition of consciousness
- 1:34:54and then two llms have no intuition and therefore llms are not conscious but wait a minute when you
- 1:35:00say felt because feeling can be ow that hurts it can be emotion and when i'll say to a lot of people
- 1:35:10i'll say oh where do you feel emotion they'll be like they'll point to their head i'm like i don't
- 1:35:14feel it there you feel it in your chest and your stomach exactly on the emotion exactly right you
- 1:35:18I think it's very telling when people point to their heads because really, and I often think LLMs are heads.
- 1:35:26They're not holistically-
- 1:35:28That's actually a fantastic-
- 1:35:29No, I agree.
- 1:35:30That's a fantastic way to put it.
- 1:35:32So the question I would put to you, though, is let's take that for granted because I think this is correct that, to extend your analogy, LLMs are just the head.
- 1:35:40They don't have the rest of the body.
- 1:35:41That is, A, that's physically true.
- 1:35:42They have no other sensory input other than tokens.
- 1:35:45And they are trained to reason.
- 1:35:47So this is basically all they are is the prefrontal cortex.
- 1:35:49They don't really have something akin to human intuition, which, by the way, is exactly why there will still be humans doing things because they don't have intuition at all.
- 1:35:58And so far, we have not at all figured out how to train intuition.
- 1:36:01Like, I don't know how you go about that at all.
- 1:36:03Well, intuition in a way, it depends what you class as intuition, because if you say intuition is just signals from your subconscious as a human being, I mean, you could easily argue that LLMs have that.
- 1:36:15And that's what's going on in that weird black box when they're deciding what's what.
- 1:36:19Because really, I think I really do think intuition is just senses that we aren't quite in touch with.
- 1:36:26And if you would just meditate and be with yourself for a long enough period of time, you would actually understand where those where those intuitions come from.
- 1:36:36They're no longer intuitions. They're just understandings of what you should do.
- 1:36:39And I think we all have different level of awareness, which for some people I'd say, you're really intuitive.
- 1:36:47And it's like, no, I'm not.
- 1:36:49You just don't understand my decision making process, which is based on a bunch of inputs that you actually don't have access to.
- 1:36:57You're either too much in your head or whatever.
- 1:37:01Because I lived with a, for example, I lived in not a monastery, but I lived with a Buddhist teacher for about a year and a half.
- 1:37:09and we meditated every day and we had i mean there's a whole rabbit hole there but the what
- 1:37:16we learned essentially was to bring to understand what awareness was and to let go of of everything
- 1:37:24because when you do let the first thing you tend to let go of is your your thoughts and this is this
- 1:37:29anchor in the in the world of of thought which is is you really do need to let go of that but
- 1:37:35There's a constant letting go.
- 1:37:37The Alan Watts chatter in the skull.
- 1:37:40Right.
- 1:37:41It's the, if you, if you, if you, if meditation is something you want to reside in, your enemy is, is what's going on up here.
- 1:37:50And it always, when you learn that, any technique of meditation, the first six months is you letting go, letting go, letting go.
- 1:38:00Especially if you were brought up in the West and you, you, you have any kind of mental acuity at all.
- 1:38:05you know but then there's there's other layers of letting go you can let go into essentially at the
- 1:38:12end you let go into awareness that's been my experience anyway which is which is why i feel
- 1:38:17like an llm i'm not sure i'm not sure i'm sure that people will plug llms into consciousness in
- 1:38:26that sense but they but as they are at the moment i can only see them as as mimic as as mimics if
- 1:38:35you see what i mean of one part of the human experience i don't know i forgot where we were
- 1:38:39going yeah no i think uh well it's a good uh it's complementary to the overall discussion of agent
- 1:38:44consciousness and what they can can and can't do so i think i would separate the two insofar as i
- 1:38:49agree with basically everything you said in your description of what this is like there's another
- 1:38:54another way to describe this by the way that i find helpful to help me kind of understand this
- 1:38:58and get more in touch with that intuitive side, for lack of a better term, is what most people in the Western world do is we identify with that little chatter in the skull,
- 1:39:09that voice that's narrating the podcast that you're listening to right now in whatever physical position you're in, that little voice in your head that's narrating,
- 1:39:16oh, well, I don't agree with what he just said, or I didn't really understand that.
- 1:39:19That voice is not you.
- 1:39:21most westerners identify with that they they they conflate this constant little narrator as being
- 1:39:28their consciousness their their conscious awareness that it is them but a fun little exercise you can
- 1:39:33do to just take a flavor of this is to in real time notice that little voice who's narrating
- 1:39:38everything that's happening right now and imagine yourself in your head try to locate physically
- 1:39:45like literally like xyz coordinates in your head where is that voice exactly try to pin it down
- 1:39:49What you'll find is you can't because that voice is not you. You can step outside of that and not identify with it.
- 1:39:55You could just observe the flow. Thoughts occur to you. They flow like a river.
- 1:40:01And you can identify that those conscious thoughts are bubbling up that that conscious stream of chatter in the skull is flowing and simply observe it Stop attaching your sense of self your identity your emotions your emotional psychological like holistic state to that thing
- 1:40:21Just simply take a step back and observe it.
- 1:40:24Just watch what it does.
- 1:40:26Watch what it says and where it comes from.
- 1:40:28And if this works for you, it doesn't work for most people.
- 1:40:30It takes years of practice and stuff, but you can get a little glimmer of it, a taste of it.
- 1:40:32If you just do that exercise for like 30 seconds and you'll see what you and I are talking about, that if you can divorce your sense of identity of the being that is you in your skull from that constant narration, that inner monologue, you'll recognize, oh, the voice is not me.
- 1:40:52The voice is a part of me.
- 1:40:55It is an active part of my prefrontal cortex of my conscious mind.
- 1:40:59But there's a lot more here.
- 1:41:00In fact, I can't locate me in my head.
- 1:41:02Like if you just try right now, locate yourself in your head, you can't because you are a holistic being.
- 1:41:07You're not a spot, a voice in your skull.
- 1:41:10Do you know what though?
- 1:41:11I exist in a completely different way then.
- 1:41:16And I think probably this stems from the fact that I've, I was taught at quite a young age, a specific way to be.
- 1:41:23But I experienced the world from here.
- 1:41:28Honestly.
- 1:41:28Oh, I believe you.
- 1:41:29Honestly, I am experiencing, I don't do that.
- 1:41:32And like the, where it comes from, for those listening, by the way, it's like, it's from the middle of my chest.
- 1:41:39It's a, it's like a, it's like a seat from where I, and when I'm listening to you, like, I just tried that experiment that you said while you were speaking.
- 1:41:49And I'm not, I'm not, I'm not thinking about what you're saying.
- 1:41:53I'm, it's just, it's just, there's like a big disc here, which is like working out what's going on.
- 1:42:00And I'm not really that obsessed with understanding what it's doing.
- 1:42:05I just know that by the end of the sentence, I'll probably know what to say next.
- 1:42:09Jesus, that sounds like an LLM, doesn't it?
- 1:42:11That's really weird.
- 1:42:12So to come back full circle, this is how the human brain does reinforcement learning through a different form of prediction.
- 1:42:18But it is akin to next.
- 1:42:19That's where we got the idea from, was studying how the brain works.
- 1:42:21I just want to finish that thought, though, because what I'm describing is something I learned, like I say, at a young age,
- 1:42:27which was a year and a half's worth of meditative practice of letting go and using the the sense of
- 1:42:37your body as the way to sort of like first let go from the thought process so you need a for want of
- 1:42:45a better word and words are really important here you need a distraction from you need and the the
- 1:42:50physical sensations in your body can be that distraction and you can train yourself to exist
- 1:42:55what you realize is quite quickly that your whole body is buzzing with vibration you just don't
- 1:43:01notice it when you're stuck in your head and the more you notice it the more you realize oh shit
- 1:43:05this is actually me when I'm non-thinking I'm not I'm not wrapped up in my thoughts you can still
- 1:43:12exist in the world you can still go about your day you don't actually need to use your brain a
- 1:43:17lot to sort of like do ones and nought stuff but what you do learn is that the best place that it's
- 1:43:23like it's a little bit like in in star wars where you you get those mecha mechatronic things there's
- 1:43:30a guy a little guy in the middle and there are these massive things that's what it feels like
- 1:43:33but you're here you're you're you're existing the actual you that's you exists here and monitor and
- 1:43:39is kind of like even monitors the wrong word because that's quite cerebral it's like it's
- 1:43:44awareness is what it is and it resides here it's weird and that awareness can be so much stronger
- 1:43:50than you think. So about like three, four months ago, I consciously made the decision to stop
- 1:43:57like actively consciously thinking through things as much, not like everything in life,
- 1:44:02some things you should think through, obviously, but as a general heuristic,
- 1:44:06this podcast is actually a perfect example. And I, it's why this is always so fun. I always
- 1:44:11intentionally do this with you, which is I'm not thinking about what I'm going to say before I say
- 1:44:15What you do is it is very akin to a flow state, right? So you completely quiet that you focus on the your general awareness of being, which quiets that little inner monologue, which has been totally absent for me as well throughout this conversation.
- 1:44:29You can turn it like on and off once you kind of learn how to do it.
- 1:44:33And instead, what that does is it makes you intensely aware of and much more open to the entire enormous – and it's far broader than you would have ever imagined – sensory like envelope of awareness.
- 1:44:46And I don't mean sensory just in the five senses, but also for want of a better term, this is going to sound like woo-woo.
- 1:44:52I don't know how else to describe it.
- 1:44:54But just the input signals from other people, from other minds, from the outside world in general.
- 1:45:01And it will make you attuned to things you could not feel or notice before, which for me occur a bit like a gut intuition from here, very similar thing.
- 1:45:11But they're much like softer and more subtle.
- 1:45:14And you can't – they're so quiet.
- 1:45:16You can't notice them if that voice is just narrating everything 24-7.
- 1:45:19It takes up all the room.
- 1:45:20It's all you can pay attention to.
- 1:45:21But if you can step back from that and expand your awareness holistically to quieten that little narrator, that little voice in your head, then you will receive effectively for want of a better term.
- 1:45:33And terminology is slippery here.
- 1:45:35for want of a better term, you will receive the impulse, the information, that subtle felt sense
- 1:45:43of what to say next, what to do next, how to connect with and interact with other people
- 1:45:49around you to a much deeper and more intensive degree. Because what you're doing is you're just
- 1:45:54simply paying more attention. You're just paying attention more. But it's not people get messed this
- 1:46:00up because they think of paying attention as like, I need to focus my mind really hard on what he
- 1:46:05No, that's not how this works at all. It's in fact doing the opposite. It is broadening your awareness to such a degree that you can pay attention to and a whole suite of inputs to the thing that is you that you are not in any way, shape or form aware of consciously normally. And even while it's happening, you're not consciously aware of it. It's just happening.
- 1:46:25and it actually one of the reasons why a lot of people are scared of it especially in a scenario
- 1:46:32like this is because you have to trust like normally there are there are some interviews
- 1:46:37like when i'm in an interview situation with someone who's not in a flow state it's very hard
- 1:46:44work you can tell right it's super hard work and it happens with politicians it happens with people
- 1:46:49who've practiced a lot who've said the same thing over and over again and it and it's really it's
- 1:46:54like it's very difficult to operate in that system and you end up with a bad interview because who is
- 1:47:00this person i'd choose not to really even talk to those people now if i can help it um but but um
- 1:47:07what i was going to say about the voice in the head thing that's a very western idea i mean i
- 1:47:13don't know whether you've traveled a lot but i lived in southeast asia a lot and i've lived in
- 1:47:17the himalayas and places like that that's the general state of being in a lot of those places
- 1:47:22you realize people are operating in the world much more like that.
- 1:47:27If you go to a village high up in the Himalayas,
- 1:47:30you'll meet people living very simple lives, or at least I did.
- 1:47:34You start to understand that that's the modus operandi of generally of people.
- 1:47:40But when you go to a lot of, you know, where we grew up in the West and stuff,
- 1:47:44you just get this obsessive brain head modus operandi
- 1:47:49of interacting with, you know, with the world through that way.
- 1:47:53But you know, go back to the flow state in a conversation.
- 1:47:58You have to be, you have to trust your understanding from here so deeply
- 1:48:04that a millisecond before I know you're about to stop speaking,
- 1:48:09I have no idea what I'm about to say.
- 1:48:11Sometimes I write something down to remind myself,
- 1:48:14but I'm already ready to sort of see it go in a different direction.
- 1:48:19And that is, I think, the natural way we're supposed to all interact together.
- 1:48:25Totally.
- 1:48:26The problem is in the Western world, it is such a deep rarity to meet someone who is self-actualized.
- 1:48:32So I think that terminology, that conceptual space is what maps onto what you and I are describing here.
- 1:48:38Because you put it perfectly in words that I have never been able to describing what that experience of doing this in a flow state feels like, which is the trust aspect.
- 1:48:48You're trusting that the deep you, the you that is intuitive, that I don't know what I'm saying next.
- 1:48:55I'm in the exact same situation.
- 1:48:57That that you is not – you have to let go of these fragile ego concerns that, oh, what if I say something low status?
- 1:49:03What if I say something that sounds silly or stupid or that is embarrassing?
- 1:49:06So if you're not a self-actualized person, if you have not holistically integrated those insecurities, that identity, that ego into a – for lack of a better term, again, it's impossible to avoid spiritual terminology here.
- 1:49:20a higher kind of state of being, of achieving a higher state of general kind of experience and
- 1:49:26awareness in life generally, then you can't – it's very hard for you to drop into that,
- 1:49:31to be in that flow state, to have that experience because you don't trust – you have too many
- 1:49:37insecurities and ego frailties and quirks, things you haven't worked out with yourself,
- 1:49:43parts of your own psyche, your id and ego have not been reconciled to, say, Freud here.
- 1:49:48Jung would say you have not integrated the shadow self, right? All of these things are different ways
- 1:49:52of thinkers trying to put into words this thing that is by its very nature difficult, if not
- 1:49:59impossible, to compress into spoken language, which is this self-actualization. But the
- 1:50:04interesting thing is, is when you come across someone, like I say, in an interview situation,
- 1:50:11that would be a politician, a very well-polished politician. It doesn't work. You don't work.
- 1:50:18And I think it's something similar to, I think someone told me this once, that in a game like Backgammon, if you're a super, super experienced player and you play a novice, it's very difficult.
- 1:50:31I experienced this with a game I play, Rocket League.
- 1:50:33Right, okay.
- 1:50:34But when you play another super experienced player, it's amazing.
- 1:50:38You know, you really get, but you can't operate at a low level.
- 1:50:43It's a bit like that.
- 1:50:44That's exactly, that's a great analogy.
- 1:50:48Yeah, I play competitive Rocket League a lot.
- 1:50:50I have a similar thing where if I drop,
- 1:50:52I'm in like the top 0.1% Grand Champion 2,
- 1:50:55which is like the second highest rank.
- 1:50:56What's it called?
- 1:50:57Rocket League.
- 1:50:57It's a video game.
- 1:50:58Car soccer, basically.
- 1:50:59Super fun.
- 1:51:00It's car soccer.
- 1:51:01It's my only like true hobby.
- 1:51:03I love it so much.
- 1:51:04Car soccer?
- 1:51:05Yeah.
- 1:51:05What does that mean?
- 1:51:06Oh, dude, it's phenomenal.
- 1:51:06Instead of players, you've got cars.
- 1:51:08You're controlling a car and the car can jump, flip,
- 1:51:11and it has rockets so you can fly in the air So there a whole like aerial aspect of the game How did you get into that Why is that a high value pastime Tell me Oh it not a high value pastime at all
- 1:51:22I don't think, I reject the notion
- 1:51:24that your pastimes have to be high value.
- 1:51:26All right, but you're the kind of gentleman that,
- 1:51:28you know, you put your mind to things.
- 1:51:30Maybe it's just you need to switch off.
- 1:51:31Well, I did that with this.
- 1:51:32I like train, I sweat, I sweat super hard.
- 1:51:34Like I trained it, I learned advanced mechanics.
- 1:51:36Like I think about the metagame and strategy.
- 1:51:39Of course, I apply the same toolkit to that.
- 1:51:41I can't not – I can't like half do anything.
- 1:51:43I just throw myself fully into whatever I'm doing.
- 1:51:46Having said that, it's a similar experience because if you play a lower-ranked player, they're not doing the things that you would – they're not doing the correct things at the times when they should be doing them.
- 1:51:56And it throws off that flow state because in the flow state, your internal intuitive sense of prediction effectively is firing on all cylinders.
- 1:52:07And when people start consistently breaking that pattern, which is you're describing in
- 1:52:11like a politician interview, they're breaking that pattern.
- 1:52:14You're not, you are aware that you're not getting to the, here's another way I put this.
- 1:52:17And again, using this terminology, cause I don't have anything better.
- 1:52:20I think this is one of those things that falls into one of my favorite categories, which is
- 1:52:23literally false, but figuratively true.
- 1:52:25So these are concepts, things that like in a deterministic materialist, physical,
- 1:52:32literal sense are false, but figuratively as a conceptual tool and mechanism to understand
- 1:52:36something are truer than the literal false definition of the thing, which is frequency.
- 1:52:42So you and I don't even know what I mean by that. I'm not going to define it. But you and I are on
- 1:52:46the same frequency every time we do this, which is why it flows. We're able to bounce around quite
- 1:52:50naturally. And we both have a blast and explore a bunch of fun topic areas and stuff because we
- 1:52:56are on the same frequency, whatever that means. I have no idea. When you're with someone who's not
- 1:53:01also able to let go and to drop into that intuitive, trusting, non-thinking sense of
- 1:53:09self where they're actively thinking about what they're going to say before they say
- 1:53:12it instead of just going like you and I are, then you can't do it because they're not
- 1:53:17on the same frequency as you.
- 1:53:18They're not on the same wavelength.
- 1:53:19It's like in an old-timey car when you had a radio with a dial where you had to tune
- 1:53:23into the station yourself.
- 1:53:25It wouldn't just seek to the right one immediately.
- 1:53:27You would turn it and you would be able to hear what the music was a little bit, but
- 1:53:31not fully and it was kind of garbled it's like that you're not on the same cognitive psychological
- 1:53:36frequency so you can't jive together because the the you're in two different strata of wavelength
- 1:53:43you and therefore the waves don't overlap it is a i mean it's a a deep truth this is why they say
- 1:53:52you know you're the sum total of the five closest people around you you know it's true it's a deep
- 1:53:57truth this is why relationships fail this is why you know um things don't work in the world
- 1:54:04is when you don't meet in the same place unfortunately correct and this extends all
- 1:54:10the way over to bring something back up that i mentioned earlier the exploration of other like
- 1:54:16what's called parapsychology so capabilities beyond the traditional cognitively psychologically
- 1:54:22the rabbit hole you went down yeah so i've been a toying with this and experimenting with it myself
- 1:54:27and it's been so fun.
- 1:54:28Hold a minute.
- 1:54:28Before, it just reminded me of something.
- 1:54:32Did you read recently?
- 1:54:33Sorry, I'm sorry to butt in there.
- 1:54:34That was a bit rude of me.
- 1:54:35But I read recently, I don't know if this is true,
- 1:54:39that somebody had trained an LLM to remote view.
- 1:54:44Is this bullshit or is that really true?
- 1:54:46He provided no proof.
- 1:54:47It was on a Joe Rogan episode.
- 1:54:48It was a philosopher, a neuroscientist.
- 1:54:52I'm not sure what his title is.
- 1:54:54It was Donald something.
- 1:54:56All right.
- 1:54:56I didn't look into it deeply, but I looked at it and I thought, that sounds interesting, but I'll deal with that another time.
- 1:55:03But it sounds like you've gone there.
- 1:55:05I did.
- 1:55:05Well, I just listened to the episode.
- 1:55:06So while I was listening, what I typically do when I'm listening to a podcast is I'll have a thread with an agent open.
- 1:55:11And I just, I tell it, hey, I'm listening to this.
- 1:55:12Go look at the episode and I'm going to get you to fact check stuff while I'm listening.
- 1:55:15And this was one of the things.
- 1:55:16And what did it come up with about that?
- 1:55:18The dude has just made the claim.
- 1:55:19He has literally no proof to anyone ever.
- 1:55:20And I don't believe him.
- 1:55:22It was an interesting concept.
- 1:55:24It wasn't something that I immediately discounted.
- 1:55:27I thought about it and I thought, well, yeah, that could be true.
- 1:55:31That could be true.
- 1:55:32And if it was true, that's really fascinating.
- 1:55:35Because, you know, to come up with something like that, that's great.
- 1:55:39Oh, it would be phenomenal.
- 1:55:40Oh, totally.
- 1:55:41But if it were true, it would be trivially easy for him to share the transcript of where it happened and prove it, like, without a doubt.
- 1:55:47So that particular – the category of wild claims that I doubt, and as you'll hear in a minute, there's a lot of them that I take quite seriously.
- 1:55:57The wild claims that I doubt are ones that are wild for which it would be very easy to provide direct irrefutable proof, and they don't do so.
- 1:56:05That's the one from, okay, why not?
- 1:56:06Like this would take you two minutes to export the transcript and send it to publish it, whatever.
- 1:56:10So that would I think would be cool, probably worth pursuing.
- 1:56:14But I – until someone publishes some sort of verifiable transcript thing where we can see that this happened, I'm putting that into the would-be-cool doubt.
- 1:56:25We'll wait to see someone else figure it out.
- 1:56:27So there's a way you can toy with these kind of parapsychological phenomena yourself though because it is intimately related to the self-actualization, the removal of the egoic self, the flow state, the intuitive side of reality, of a more holistic being.
- 1:56:42which is what it like is typically traditionally described as is listening to your gut more
- 1:56:48but what that really means is to do this you really do have to shut your conscious mind up
- 1:56:54almost completely and just feel like to to be aware of and feel your whole being and then notice
- 1:57:01the little tiny like gradients and stuff that are there so i've been toying with this in a variety
- 1:57:08of ways after I went deep down the rabbit hole and like how people do this. And there's a bunch
- 1:57:12of research on it, turns out. So there's a body of research. Anybody listening, go ask your clinker
- 1:57:17about. There's a large body of research where they do these experiments of they'll put a number
- 1:57:22or a shape in an envelope, put the envelope in another room, put a person in a firewalled room.
- 1:57:28There's no possible way the information to cross. They have the person guess what's in the envelope
- 1:57:31given a quite broad range of possibilities. It could be these 20 shapes, these 26 letters,
- 1:57:37or these, you know, hundred numbers.
- 1:57:39So the combinatorical math there is like there's more possibilities
- 1:57:42than there are atoms in the universe.
- 1:57:44So-
- 1:57:44Are there any inputs at all?
- 1:57:46They just say there's a number in there.
- 1:57:48There's a thing in there.
- 1:57:49And here's like the-
- 1:57:50A thing, not even a number.
- 1:57:51They only narrow it a little bit,
- 1:57:53enough for the math to still be provable,
- 1:57:55that it's like super low probability.
- 1:57:57They would just guess it right multiple times.
- 1:57:59And it turns out there are individual studies
- 1:58:02that showed statistically significant predictive abilities
- 1:58:04and others that showed negative-
- 1:58:06Is there a technique to try and work it out?
- 1:58:11Yes.
- 1:58:11Or do they just say anything, just say a number?
- 1:58:13So they have found, so recently, in the last five years or so, a meta-analysis of these studies was done.
- 1:58:20And they found of people who claimed before the study to be able to do this was like 60 to 65% more hit rate, which is way, it should be 50-50 on the nose.
- 1:58:30Because it's a coin flip, basically, or lower, depending on the study makeup.
- 1:58:34It was way higher than chance, is my point.
- 1:58:35Like several orders of magnitude higher than chance.
- 1:58:39And there's a trick to this, the way they do it in a nutshell.
- 1:58:42And again, oversimplifying here, like this takes a lot of like practice and kind of intuiting
- 1:58:47and just experimenting with it, is you have to shut that conscious mind up completely.
- 1:58:51And just whatever, it'll feel like a guess.
- 1:58:56Whatever answer just occurs to you when you're, that voice is silent, your conscious mind is
- 1:59:00off.
- 1:59:01That's the one you should put forward.
- 1:59:03And it may or may not be accurate, but that is in a super simplified dumbed down version how these people who claim to be able to do this to have telepathy, telekinesis, or whatever, a bunch of parapsychological phenomenon.
- 1:59:20That's how they do it.
- 1:59:21So I've been experimenting with this just in general in life, which is like anytime I sense the slightest little gut intuition toward – I just do it.
- 1:59:29If you think about it, you will screw it up immediately.
- 1:59:32Just do it.
- 1:59:33And it's been super fun.
- 1:59:35I agree with that.
- 1:59:36I feel like I've gotten better outcomes at work as a result of this.
- 1:59:40So like this is going to sound crazy.
- 1:59:41When I'm having super high stakes meetings with high level government officials or something that we finally managed to get to to try to tell them about Prospera and what we're up to and whatever.
- 1:59:50In those meetings, I am never thinking.
- 1:59:52There's not a single thought in my head the entire time.
- 1:59:54I'm just whatever my intuition says I should.
- 1:59:57Like speak.
- 1:59:57Don't speak.
- 1:59:58Say this.
- 1:59:59Don't say this.
- 1:59:59noticing you'll, you'll start, you'll find you just start picking up intuitively. This never
- 2:00:04will enter the conscious, like that little voice will never narrate. It'll never enter conscious
- 2:00:07awareness that, oh, this guy's not receiving this very well. Or he clearly like wants to push back,
- 2:00:12but is not for X, Y, Z, or like these people are very nervous and I'm, I can realize it.
- 2:00:16You'll pick up on a lot more. And then if you'll just listen to that little, those tiny little
- 2:00:22intuitive pulls and just follow them, just don't fight it. Just go down the gradient that is already
- 2:00:27pulling you in that direction, it's, for lack of a better term, magical. And I'll give you a silly
- 2:00:33example, okay? And by the way, for the, you know, materialist pedants listening, I am well aware my
- 2:00:38sample size here is like four. It's not going to scientifically reproduce. I'm sure the next time
- 2:00:43I try to do this, it'll miss, blah, blah, blah, caveats aside. One way I've been doing this,
- 2:00:46just to like experiment with it, is whenever I go to the airport, at my hometown airport,
- 2:00:51there's like this specific deck of a parking garage I like to park on every time because I
- 2:00:55I fly every week, like usually one or two, sometimes three times a week.
- 2:00:58Now, what I'm sort of doing is before I go, when I get closer to the airport, I do not think about like there's 10 lanes total, right, of parking.
- 2:01:06I don't ever think about where to go because there's almost never any parking spots.
- 2:01:11And I have to go to this like other deck that I don't like.
- 2:01:13It's a longer walk.
- 2:01:14So now when I get to the parking garage, I don't think at all.
- 2:01:17I just go where my hands want to take the steering wheel.
- 2:01:20And the past like I think four times I've done this, there's been a spot on that row.
- 2:01:24just one, sometimes two, but there's been a spot on that row. And there weren't any on the other
- 2:01:27rows. I can see as I'm driving down them, I can see that they're not there. Little stuff like that.
- 2:01:31And again, I'm sure next time I do it, it'll miss. I'm not saying this is going to give you
- 2:01:34perfect clairvoyance or whatever. Just, I do encourage people though, don't dismiss this
- 2:01:38out of hand. Like start playing with stuff like this. You'll, I think you'll be surprised by the
- 2:01:42results. If here's the critical piece though, this is so hard to, there's so many things in life like
- 2:01:47this I've found, which is that if you don't full sin commit to them, like to your core,
- 2:01:54if you're like, all right, I'm going to instrumentally, faith has been like this for
- 2:01:57me. So I became a born again, Christian, like three, two, three years ago, something like that.
- 2:02:03After a lot of like deliberation and study, how do you talk about that separately if you want to?
- 2:02:06Cause it was hilariously, the last book I read before I did was this book called Buddhism is
- 2:02:09true by this guy with a PhD who became a Buddhist and then made a secular case for Buddhism.
- 2:02:14Great book, by the way, I highly encourage everyone to read it.
- 2:02:17It improve your life no matter what your belief background is And since then I started doing this like a lot more But the problem is if you try to instrumentally because I tried to do this for years if you try to instrumentally
- 2:02:30adopt a faith, for me, it was Christianity, because I can objectively see if I were to follow
- 2:02:34these ethical guidelines and tenets, my life would be better. And churches are a pro-social,
- 2:02:39informal social institution that increase community cohesion and trust, which reduces
- 2:02:43transaction costs and causes economic growth to accelerate. You can have all these empirical
- 2:02:47arguments for it and say, okay, I'm going to do all of the move, go through the motions. I'll,
- 2:02:52I'll read scripture. I'll go to church. I'll participate in, in the church community.
- 2:02:56Instrumentally, nothing will happen and you'll be frustrated and your life's not going to really
- 2:03:00improve and nothing's going to change. But the second you like full sin commit to it,
- 2:03:03and this parapsychology experimentation is the same way. If you're not fully to your core saying,
- 2:03:09okay, I am going to actually give this a shot and I'm not going to be in the back of my head,
- 2:03:14doubting it the whole time. And I'm aware I'm doing it instrumentally. I'm just going to
- 2:03:17fully to my core as an integrated self-actualized being commit to doing this, you will have just
- 2:03:24astonishing results. The same thing is true for anything you do in life, by the way, that little
- 2:03:27nagging self-doubt when people do stuff instrumentally. And by that, I just mean they're
- 2:03:33doing it because they have rationally thought through and concluded it will result in positive
- 2:03:37outcomes in their life. This is not always the case, but you should work out. If you've reasoned
- 2:03:40your way to, I should be in good shape. Yeah, you should probably be in good. That's a good idea.
- 2:03:43This is not a generalizable, always-on heuristic. But as a general principle, if you will actually, to your deep emotional core, in your truest self of selves, say, no, I'm not going to instrumentally adopt a Christian. I'm going to believe Christ died and rose from the grave for forgiveness for my sins because the creator of the universe said so.
- 2:04:04totally different transformative experience same thing goes for this little parapsychology
- 2:04:10experimentation i've i've been doing with little stuff like this you have to actually not doubt it
- 2:04:15you have to actually be okay this is this is going to work and i'm going to give this a fair
- 2:04:18shot and i think again words fail but i think everyone knows intuitively what i'm referring to
- 2:04:23well i maybe but but you have to experience it some people do block a lot of that stuff out
- 2:04:29I have exactly the same, but in a slightly different way.
- 2:04:33And it began for me when I was younger, because I was quite shy.
- 2:04:37And I noticed urges to talk to people, which I would shut down immediately.
- 2:04:44And the more you notice them, the more you realize, okay, I'm shutting down that.
- 2:04:49What's that all about?
- 2:04:50Why am I doing that?
- 2:04:51Right, exactly.
- 2:04:53What that then transformed into after many years is waking up every morning and realizing
- 2:04:57life is an adventure and anything literally anything could happen to you today anything
- 2:05:04as long as you choose to to follow those urges and and when you have a good morning like that
- 2:05:11you go out to get your morning coffee and you bump into someone and you say hey how you doing
- 2:05:17and it changes the course of your life quite literally it's like sliding doors that film
- 2:05:23It's like you walk through, you go through the doors and your life is never the same again.
- 2:05:28And that is in every moment.
- 2:05:30There are, there's the opportunity to allow that intention to arise out of you and to pursue it and to, and to do it, to be it, to do it.
- 2:05:41And, and that is it.
- 2:05:43I mean, it's, it's, yeah, I forget about that a lot, but when I remember it, it's like, wow, I'm really vibing today.
- 2:05:50Yep.
- 2:05:50I'm really vibing today.
- 2:05:51This is so much fun, being in the world.
- 2:05:56Oh, yeah. Look, we are a – in the cosmic timescale, our lives are less than a blink of a blink of an eye on a giant spinning rock floating around a star that is in and of itself one of billions of stars in a galaxy in a cluster of, as far as we can tell, billions of galaxies that extends who knows how far.
- 2:06:21like that is patently absurd already by itself like the fact that you person who is taking life
- 2:06:27so seriously and hating it or living that way is an affront to the fact that you have been given
- 2:06:33the precious gift of existence itself like kaboo was correct existence itself is absurd and it is
- 2:06:39up to you what you do with that absurdity you can fight it you can lament it you can you know try to
- 2:06:45not acknowledge it but it is true and as soon as you stop fighting it and just say yeah life is
- 2:06:50absurd. Every moment of existence is precious and I am going to value existence qua existence.
- 2:06:56I'm going to value my life for the sake of the fact that I have been given the precious gift of
- 2:07:00happening to have the atoms come together to form me now in this time with conscious awareness.
- 2:07:08That is an unbelievably rare gift, a miracle that we owe it to ourselves, to reality,
- 2:07:15to take full advantage of, to acknowledge and to live acknowledging that that is true,
- 2:07:21that this life is insanely low probability and rare and short and precious and amazing
- 2:07:29if you would only open yourself up to the opportunity to even realize it in the first
- 2:07:33place. There's a great book about this called Greatness Cannot Be Planned by a couple of
- 2:07:36software engineers. I was going to add something to that, which is the fact that I agree with that
- 2:07:42100% and I think there's another element to that which is you create your own reality and I would
- 2:07:49almost go I mean I can't never prove this but I would almost my experience that actually is I'm
- 2:07:57projecting reality out you know because because everything I come across here is is almost like
- 2:08:03tailor-made for me it's showing me it's showing me xyz it's testing me it's allowing me to it's
- 2:08:10that literally the universe is a mirror i know it's a cheesy meme but it literally is
- 2:08:14and when you add that to life is precious you get an amazing life and you realize that that this
- 2:08:24is a reflection of the fact that you understand that life is precious and if if if shit's going
- 2:08:31down in your life pay attention to it because it you're not a victim it's it's it's a part of you
- 2:08:38it's who you are and it's it's there to be addressed it's there to be to be integrated
- 2:08:45it's there to be you know whatever and that us I have that experience so often and it's it's a
- 2:08:52because I have a really enjoyable life on the whole and off and but then when something when
- 2:08:57a blip appears it's so obvious that it needs a dress it's something that needs addressing in me
- 2:09:03you know some annoying thing happens or some person I don't like does something really stupid
- 2:09:08or causing this flow to sort of like ebb when it shouldn't.
- 2:09:14And, you know, there's never an opportunity to say,
- 2:09:18oh, poor you, that person's doing something bad to you.
- 2:09:22It's like it goes hand in hand with taking extreme responsibility
- 2:09:26for absolutely everything in the outside world,
- 2:09:29which sounds a bit weird to a lot of people,
- 2:09:31but it's the most liberating thing you can do.
- 2:09:34You know, life is sweet and nothing, well, bad things happen to you, but you own them.
- 2:09:45Exactly.
- 2:09:46And even when the bad things happen, right?
- 2:09:48For me, the momentary like emotional negative valence that occurs with that is so short and fleeting because it always just immediately reminds me of two things.
- 2:09:57A, this is great.
- 2:09:59I've learned something.
- 2:10:01I've been corrected.
- 2:10:02I was going down a wrong path somehow.
- 2:10:03The thing I was doing that went wrong, even if it appears, if you're a third party looking at an objective, like you had nothing to do with that.
- 2:10:10Well, no, I guarantee you somewhere along the line I did.
- 2:10:13And it is better for me to recognize this as a learning opportunity to say, I failed.
- 2:10:18Great.
- 2:10:18I know what not to do now.
- 2:10:20Do something different next time in whatever this arena, this thing is.
- 2:10:24And then the second thing, which is a more profound insight that, again, sounds banal unless you really kind of sit with it and ingest it.
- 2:10:31It is impossible.
- 2:10:32And I'm sure you've probably found something akin to this. It is next to impossible to get stuck with that negative valence, that negative emotion because the good experience of life, the magical, flowy, subjective well-being we're describing needs a referent.
- 2:10:57It is at its core not an absolute value.
- 2:11:00It's a referential value.
- 2:11:02So when you're in the depth of your deepest despair, you lost your job, your credit card bills are racking up, you've done 1,000 applications and haven't gotten an email back or a call back, A, it's a sign you should do something different.
- 2:11:12And B, when this turns around, and as long as you're a responsible adult at some point, it will, you can now value so much more deeply and experience so much more deeply the joy of a life well lived only because you experience the depths of pain and suffering and despair so that you have a referent to which to refer to be able to effectively appreciate and value.
- 2:11:42the wondrous experience you're currently having.
- 2:11:45So even when something terrible happens,
- 2:11:48I know, well, if nothing else,
- 2:11:51this makes me value the beauty of life so much more.
- 2:11:54I think also at the core of that is meaning
- 2:11:59because meaning allows you to navigate any experience,
- 2:12:05right or wrong, painful or joyful, whatever.
- 2:12:10it carries you through everything you know like the worst if you're you know like the great the
- 2:12:17best example is having children they cause you the dark the deepest pain and the most joy
- 2:12:23but it doesn't matter they're one and the same in that moment the the feeling that the the pain you
- 2:12:29feel thinking that your child is unsafe is no different it's on this it's it's like this it's
- 2:12:36on the same spectrum i think is the the horseshoe yeah as as the joy all it is is a connection
- 2:12:44opening up and and that and and that's what you you get that you know and and without meaning
- 2:12:50i think you chase the good vibes because you know like me
- 2:12:55happiness and and living happy is not always the appropriate scenario like when you're when you
- 2:13:07want to someone close to you dies happiness is not the correct there's something deeply wrong
- 2:13:12right exactly we but we live in a society that's got to this point where we chase
- 2:13:16We expect the meme is chase happiness. And actually, I think by by by understanding the meaning that drives you, which possibly can't prove it.
- 2:13:31arises in those moments and and your little experiment and my little experiment with just
- 2:13:37reaching out to people is meaning as well there's a meaning why else would this strange urge arise
- 2:13:45to make a micro movement in a direction what what is that what is that that's i'd call that
- 2:13:52that's actually a lot of that i'd call enthusiasm as well which is a very poignant way of describing
- 2:13:58because it's got theos in it.
- 2:14:00It really has, inspired by, I think it's theos.
- 2:14:04But it's definitely connected to the gods.
- 2:14:07To the divine.
- 2:14:08Yeah, exactly.
- 2:14:09And it is something that is necessarily tempting you
- 2:14:14to be divine, I suppose, in that sense.
- 2:14:17Yeah.
- 2:14:17It's experiencing that little kind of spark of the divine.
- 2:14:23And I think where a lot of people also go wrong is,
- 2:14:25so first of all, on your happiness point,
- 2:14:27I completely agree. And the way I conceptually separate these in my head,
- 2:14:32happiness, I associate with like hedonistic pleasure, the pleasure of the senses, right?
- 2:14:38To contrast with joy, joy, you can feel joy even in the darkest of moments because joy is
- 2:14:46associated with this deeper existential sense of meaning and of gratitude and of appreciation
- 2:14:51for existence itself. That is where joy resides and in gratitude for existence,
- 2:14:57whatever your existence may be at the time, good, bad, ugly, whatever.
- 2:15:01Happiness is the brief fleeting chase of the pleasure of the five senses. It is dopamine
- 2:15:07maxing effectively, which is why I think modern culture is so deeply, deeply broken
- 2:15:12is this is actually at the absolute core of it. I'll sit with you. I just wrote an essay about
- 2:15:17this on my personal site called the Good Heart Civilization, where it's kind of my theory of
- 2:15:22everything for why everything sucks now, because most people think most things suck now just in
- 2:15:28general. And what has happened is the market mechanism used to incentivize pro-social behavior.
- 2:15:34You make a profit by creating things people value writ large, which as a whole was a coarse proxy
- 2:15:40variable, was correlated with a holistic sense of well-being, of human flourishing broadly
- 2:15:45construed, right? Well, now, thanks to advances in technology, specifically software, machine
- 2:15:50learning, things like that, you can just skip the proxy variable. You don't have to create a thing
- 2:15:54that creates holistic value in people. You can just optimize directly on the underlying biological
- 2:15:59dopaminergic reward system itself. You can gradient descent the reward function in the human brain
- 2:16:04itself, which is why there's a great study that I cited in the paper that drives this point home
- 2:16:09better than anything I ever could, which is they asked a bunch of college students how much someone
- 2:16:14would have to pay them to stop various social media sites, and they all gave various values.
- 2:16:17But then they asked them in a really interesting question, how much would you pay to have this
- 2:16:23thing deleted from existence? No one's on it. It's gone. It ceases to exist as a whole. So you
- 2:16:28and no one else are on it. And those also had massive positive values associated with them.
- 2:16:34You would think it'd be zero or negative, because if you're enjoying TikTok or whatever,
- 2:16:37TikTok was the one people would pay the most to stop, by the way, because it's the one that most
- 2:16:41directly hijacks the dopaminergic reward system of the human mind. The algorithm does so. And this
- 2:16:46basically is happening to everything, right? Because we lack a coordination mechanism to get
- 2:16:51everyone to stop these things. And so every industry is now like this. It's value capture
- 2:16:55and value extraction rather than value creation. And it is optimizing against the biological reward
- 2:17:02system itself. It's like they find the perfect suffering Lagrangian point. So a Lagrangian point
- 2:17:07is where the gravity between two stellar objects nets to zero. So you can just sit perfectly
- 2:17:11stationary in space. It's like they find a perfect suffering Lagrangian point exactly where the thing
- 2:17:17kind of sucks now, but it doesn't quite suck bad enough for you and everyone else to stop using it,
- 2:17:21which would be a market signal for them to stop doing that. So companies are now optimizing to
- 2:17:24find the suffering Lagrange point. It's why everything sucks. Coming back to the full circle,
- 2:17:29your happiness point though, that's why this is happening because the entire market mechanism now,
- 2:17:33the machine, there's a great book called The Machine Against God. I encourage everyone to
- 2:17:36It's half philosophical treatise, half rant.
- 2:17:38It's phenomenal.
- 2:17:39It's a similar point from a different vector.
- 2:17:41And the reward system of the market mechanism itself is now rewarding this hacking of the human dopaminergic reward system, happiness in a nutshell, which is – that's what happiness is, hedonistic pleasure.
- 2:17:51and human holistic well-being, human flourishing broadly construed at an existential spiritual
- 2:17:57level is deeply suffering because all of society is oriented toward chasing happiness, chasing
- 2:18:03dopamine, chasing sensory pleasure and reward, and not the deeper existential values of human
- 2:18:09flourishing, which can only come from gratitude and appreciation for existence and a sense of
- 2:18:14meaning and purpose that are driving you toward what you're doing. Because every job, this is
- 2:18:19where the whole, like, I think bullshit jobs category comes from. The whole thing is fake
- 2:18:22itself. Like there's, there's, there are no bullshit jobs because the first firm to just
- 2:18:26fire the bullshit people would obviously be the most prosperous one. So that, like, that is not
- 2:18:30true. But what is true is modern jobs, especially white collar jobs that a lot of people work that
- 2:18:34are on social media suck the soul out of life. You're just, there's, there's, it's so divorced
- 2:18:39from meaning and purpose and true value creation that I can see where people, this idea of bullshit
- 2:18:46jobs resonates with people. It's not literally true, but it's figuratively true. These jobs are
- 2:18:49bullshit in the sense that they do not provide you any purpose or meaning. They give you money
- 2:18:54in exchange for suffering so that you can chase further hedonistic pleasures. I think this thing
- 2:18:59you're describing, this focus on happiness rather than joy and meaning and purpose is reflected in
- 2:19:06the actual mechanisms that companies have figured out that you can optimize on the human dopaminergic
- 2:19:11reward system instead of human value creation and human flourishing itself. It's the same
- 2:19:15phenomena viewed from two different analytical lens. And it is the root of a lot of modernity
- 2:19:20suffering. It is the root of many of these problems. I think a lot of Bitcoiners would
- 2:19:25argue that that is a function of fiat money as well. Short-term thinking, chasing bullshit,
- 2:19:36food being bullshit. Same thing, hijacking your dopamine. Everything. Everything, yeah.
- 2:19:42It's funny enough, I told you I'm doing the last speech,
- 2:19:50or the second to last speech at this conference.
- 2:19:52That's what I'm doing.
- 2:19:54I know, but it wasn't, I'm not trying to point that out.
- 2:19:59The joke is all the talks are 15 minutes, you know that.
- 2:20:03And we've been going two hours, 20 minutes, right?
- 2:20:08This is proper.
- 2:20:09This is how you should interact.
- 2:20:11Right, right.
- 2:20:12Not a 15-minute talk.
- 2:20:14So the joke is, this is a 15-minute talk.
- 2:20:16So I've just created a load of AI slop.
- 2:20:19I can't wait to see this.
- 2:20:21Well, I hope it lands well.
- 2:20:23I just couldn't get up there and talk about free cities
- 2:20:27without just giving something really ridiculous.
- 2:20:30But there's a chance it might not land.
- 2:20:33It might not.
- 2:20:34We'll see.
- 2:20:35Yeah, well, I cannot wait to see this.
- 2:20:38And on your Bitcoiner point, by the way,
- 2:20:40this is one complaint.
- 2:20:41There's a whole section of this essay just attacking libertarians.
- 2:20:45And I say this as someone who like Rothbard changed my life.
- 2:20:48Like I was a dyed-in-the-wool anarcho-capitalist for a long time.
- 2:20:51Bitcoiners suffer from the same thing libertarians do.
- 2:20:53It is this – I could put this in.
- 2:20:57There was a viral tweet from a mutual of mine eigenrobot that described it perfectly.
- 2:21:01All of the problems in society are caused by whatever topic I happen to be obsessed with this given month.
- 2:21:07Bitcoiners, a bunch of other people, Bitcoiners, it's fiat currency.
- 2:21:10libertarians as the state. There's a bunch of different, like, yes, these are obviously
- 2:21:13problems. Like the Bitcoin or critique of the fiat currency and constant inflation causing
- 2:21:19short-term rather than long-term incentives and reducing time preference. It's all true.
- 2:21:24And there's a bunch of other things going on causing the same things. Like the universe,
- 2:21:28the world is incredibly complex, multivariate. All of these things are recursive feedback loops
- 2:21:33in and of themselves. It's impossible to disentangle them. So yes, Bitcoiners, you're right.
- 2:21:38and I'm right.
- 2:21:39And you have a different theory that's probably right.
- 2:21:41But I think probably the truism there
- 2:21:45is that underlying it all is some incentive model
- 2:21:51that causes you to search for the short term.
- 2:21:55And even fiat currency is a function of that.
- 2:21:59It is the reason why,
- 2:22:01but it's not right at the base of it all.
- 2:22:04It's effect and cause.
- 2:22:05It's a recursive feedback loop.
- 2:22:07And it's probably, when you think about it, there's probably a binary there at the bottom.
- 2:22:14And there probably is an either or.
- 2:22:17You're either a bit like you're either a collectivist or you're an individualist.
- 2:22:21I don't know what it is at the moment, but there's probably something deriving it all, actually.
- 2:22:26I think it's a spiritual, like existential sickness, which is...
- 2:22:32But why?
- 2:22:33But why would you have that?
- 2:22:35like that's like that's like saying right so there's you know there's the guttural
- 2:22:40desires and then there's the spiritual experience and that's the binary and and this is one the
- 2:22:49spiritual experience is long term it's hard money it's it's it's renaissance that the other stuff is
- 2:22:56yolo short-term thinking lust you know all those things that dopamine blah blah but but what's the
- 2:23:04underlying incentive model what is that thing that's causing either one or other to happen
- 2:23:11uh the i mean why do we do why do we do things why do we respond yeah i mean yeah why is there
- 2:23:20a difference why do we respond to sexual urges on one hand and nurturing children for 25 years
- 2:23:27they're both part of the same process right yeah the gulag archipelago the uh line between good
- 2:23:35and evil cuts through the heart of every man i think we have so like i'll put this in both
- 2:23:40spiritual and secular terms uh man's man has a nature by default but what nature is expressed
- 2:23:48or suppressed the the instantiation of your nature is shaped by incentive gradients everywhere and
- 2:23:56almost every facet of life, in the personal, the spiritual, health, financial, whatever.
- 2:24:03And right now, all of those gradients effectively kind of push the same way. To put it in spiritual
- 2:24:09terms, man has fallen and sinful, and the degree to which one is either attempting to follow
- 2:24:17closely in Christ's footsteps to try to live a life closer to the Lord in every sense of that
- 2:24:21that term, or giving into hedonistic pleasure and giving into the incentive gradients as
- 2:24:26they exist.
- 2:24:27This line, framed in these two totally different ways, is the same line that cuts through the
- 2:24:33core of every man and every woman I saying man in the philosophical sense here So i think the culture the society the circumstances in which you live determine which side of that that yin yang you know again
- 2:24:48there's uh i cannot i always come back to the deep truths that cut across spiritual traditions i find
- 2:24:53the fact this is one of them this this concept of the yin and yang right i can see the symbol in my
- 2:24:56that's why i drew it like oh that's hilarious no no but it's the it's what i like about the yin
- 2:25:02and the yang is it's not a straight line and it has the option to put the the same thing in the
- 2:25:07middle of on the other side as well which is really important exactly exactly i love it and that that
- 2:25:11is in effect a description of man's nature the christian would call this the fallen man trying to
- 2:25:17you know pursue god the buddhist or the the taoist whatever would you know describe it in the
- 2:25:23traditional you know terms of the the yin and the yang of the tao it's the same exact thing just
- 2:25:28carried across completely different cultural contexts. And the deep truth that I have come
- 2:25:33to realize over the last – this has happened to me over the past couple of years probably,
- 2:25:36kind of part and parcel of this whole kind of self-actualization journey, has been recognizing
- 2:25:42the deep, deep importance of culture writ large. And I mean that in the broadest sense of the term
- 2:25:47culture. When people think of culture, they typically think of like entertainment artifacts.
- 2:25:49I mean it much more deeply, like values, norms, social expectations, the deep sense of culture
- 2:25:54is I think at this point, a more important incentive gradient in some deep sense than
- 2:26:03even the more explicit, measurable, rational, calculatable incentive gradients that economists
- 2:26:09study, that social scientists study. The cultural gradient just gets more – I think it gets less
- 2:26:15discussion and it's harder to discuss because you can't point to empirical studies numbers.
- 2:26:19You can't quantify it anywhere nearly as easily if at all. But I think those cultural norms and
- 2:26:23values. Deep culture determines the wealth and prosperity, the fate of a people. And again,
- 2:26:30there's intense recursive feedback loops here, right? The economic hard incentives shape culture.
- 2:26:37Culture in turn shapes how people behave inside of those explicit formal institutional incentive
- 2:26:43gradients, which in turn further shapes the culture. These things are completely inexplicable
- 2:26:47and impossible to disentangle. But I think that we, as moderns, the deep thing that's happening
- 2:26:55there that I think you and I are both kind of getting at is this spiritual, cultural rot,
- 2:27:00the culture. Because you can imagine two societies that have the exact same material
- 2:27:04experience, like snapshot in time of modernity. But one of them, let's say, and to abstract it
- 2:27:11away from like controversial, like religious points, let's say one of them is like purely,
- 2:27:14perfectly Buddhist. It is embodying the Buddhist spiritual and ethical tenets maximally in the
- 2:27:22exact same circumstances we are. And then you press play on those two worlds. So the world as it exists
- 2:27:27in this Buddhist world, right? Buddhist America. The outcomes immediately diverge, even though the
- 2:27:32starting point, the starting condition was the exact same formal institutions and therefore
- 2:27:35formal incentive gradients, but they immediately radically diverge. So I think culture ultimately,
- 2:27:41what that thought experiment shows to me, at least when I'm thinking through it,
- 2:27:43is that culture is a deeper, truer thing. It is a deeper issue that if you can fix,
- 2:27:51heal somehow, change the culture itself, then it becomes – it's not a given at all,
- 2:27:56to be clear. But it becomes easier to downstream change the formal institutions of governance,
- 2:28:03of law, of economics, of business, of capitalism, whatever. You can change the formal institutions
- 2:28:08much more easily if you have deep alignment on the cultural institutions. That's why the Amish
- 2:28:12exist, right? They have deep alignment on iconoclastic cultural ideas, which has let them
- 2:28:18create a system, a set of formal institutions that look nothing like the rest of the world.
- 2:28:24Existing within the rest of the world, by the way, it's not like they're, you know,
- 2:28:27totally on a mountaintop somewhere that don't do any economic interaction. They're economically
- 2:28:32active. But because their cultural institutions are so fundamentally different, the spiritual
- 2:28:37cultural institutions, the formal institutional outcomes are much, much different.
- 2:28:41All right, then. I've got a question for you. Since you could easily say that Prospera is a version of that, what's the culture of Prospera? Or what should it be?
- 2:28:56What it should be in my ideal world is a culture maximally on the opposite end of the time preference scale from the rest of modernity.
- 2:29:07It should be a long-termist culture oriented toward human flourishing, toward value creation, a culture that intentionally crafts itself toward the ends of deeper human flourishing, not just how much money can I make per se.
- 2:29:22That is but one aspect of a much broader holistic view of life.
- 2:29:25It should be one in which individual expression, autonomy, creativity, expressing the truest, deepest sense of oneself, whatever that may be without regard for the traditional kind of social signaling and social status mechanisms that keep people trapped in these eco-defense mechanisms that keep them from self-actualizing.
- 2:29:44In a sentence, it would be a place where self-actualized people go and where people on the journey to self-actualization come.
- 2:29:52That would be what I would love to see.
- 2:29:53I'm not at all saying it's that.
- 2:29:54Yeah, but that's what I would love to see.
- 2:29:56Sure.
- 2:29:57Imagine going to a board meeting and laying that out, though.
- 2:30:01It's a tough one.
- 2:30:02The next question is, yeah, but we need a new coffee shop first.
- 2:30:06And then the question after that is, okay, that all sounds great.
- 2:30:08What are you doing tomorrow?
- 2:30:11How are you doing that?
- 2:30:12No, but I mean, what you're describing there is the ebb and flow or possibly the yin and the yang or the wheel of life.
- 2:30:20And arguably, prosper is inevitable in that sense.
- 2:30:24Because all it is, is a response to the other side.
- 2:30:29You know, what's happening?
- 2:30:32In fact, what's happening here is the dot in the yin and the yang.
- 2:30:38When you draw the yin and the yang,
- 2:30:40prosper is one of the dots on the other side.
- 2:30:43It really is.
- 2:30:44And it exists because it creates balance.
- 2:30:48I mean, this sounds really retarded and stupid, but hear me out.
- 2:30:53It's important.
- 2:30:54And it will happen.
- 2:30:55Or if it doesn't happen here, it'll happen somewhere else because the cycle of life requires it.
- 2:31:02If you go when the short term thinking becomes so short, the response is long term thinking.
- 2:31:10And that's the yin and the yang.
- 2:31:12And the two dots are the points where the old emerges into the new, which is what prosper is, basically.
- 2:31:20That's exactly right.
- 2:31:20You see this over the long scale of like human civilization time horizon.
- 2:31:24If we look at ancient Egypt, ancient Inca, Maya, these very old civilizations that were large centralized, which then ended up devolving into decentralized kind of clans, tribes, et cetera.
- 2:31:38re-centralizes under the kings and the royal authority, which then decentralizes again into
- 2:31:45feudalism, re-centralizes into the modern state, and then a competitive jurisdiction world, a world
- 2:31:50with a bunch of decentralized jurisdictions is obviously what's coming. It's the same
- 2:31:54centralization-decentralization cycle that humanity has always gone through.
- 2:31:58And what's funny is I love when this happens, but I keep finding these synchronicities between these
- 2:32:02ideas. The way we're describing it, the frame through which we're describing this right now
- 2:32:06It's very speculative, kind of spiritual, cultural out there.
- 2:32:10But you can also just describe this exact same phenomena in hard empirical terms, which is currently the dopamaxing, I call it, the dopamaxing people are not reproducing demonstrably.
- 2:32:21The world total fertility finally across the whole human population dropped below the replacement rate.
- 2:32:25The fertility crisis is a huge problem.
- 2:32:26It's a whole deep rabbit hole I can get into.
- 2:32:28But it's a huge existential crisis.
- 2:32:31But the people that are above replacement rate are the long-term thing.
- 2:32:34It's the Mormons.
- 2:32:35It's the Amish.
- 2:32:35It's the Catholic. It's the highest earners who can afford to have a bunch of kids.
- 2:32:41So you end up with – and these people in turn, the way they became high earners is by like long-term thinking, delayed gratification, et cetera.
- 2:32:47So if you just play the biological clock out, you are rapidly selecting for the type of people who reproduce are the ones with the long-termist thinking, the ones with the delayed gratification, the hard money people, if you will.
- 2:33:00so play it out a few generations and that's the only people left because the short-termist people
- 2:33:05the the as i call them the gay race communists don't reproduce but why why does the long term
- 2:33:12the why does the the low time preference group give way to the high time preference group i think
- 2:33:20it's a matter of like short-term incentives it is uh but why do they get why do they get over
- 2:33:26powered by them like how do we end up in this situation when quite obviously long the long-term
- 2:33:33let's call i don't know what we call it what's the opposite of the dopamaxers that's a good
- 2:33:38question uh whoever that is yeah right think of a name whoever they are they seem that how do they
- 2:33:46get out competed by the bullshit dopamaxers i'll answer this by way of analogy imagine two
- 2:33:52identical firms that are publicly traded. Firm A has a low time preference, long time horizon
- 2:33:59executive team board. So they are investing their free cash flow heavily into developing new
- 2:34:05products, new business lines, et cetera. Firm B is in, imagine they're clones. They're identical
- 2:34:10in every way, except firm B is purely maximizing quarterly profit. Well, on any given quarter,
- 2:34:17firm B wins every single time. So where does capital flow? Firm B, because, well, they're
- 2:34:21quarterly earnings printed. And firm A's did not at all because they're investing heavily in the
- 2:34:26development of new things that are going to create way more value down the line, three, four, five,
- 2:34:30six, eight quarter from now. But capital flows to the firm, which is maximizing quarterly profit
- 2:34:36right now. The entire private equity industry is effectively just taking this to the nth degree,
- 2:34:41right? I have a whole section in that essay about how private equity is the scourge of modern
- 2:34:44civilization. Not the only one, but one of them by kind of going through this in great detail.
- 2:34:49But the point is just there's – in any given short time horizon, arbitrarily constructed, the short-termist people win.
- 2:34:57Now, over the long-term horizon, longer, the longer-term thinking people win.
- 2:35:02But you only need to appeal to the short-term thinking people and therefore change the formal institutions to more benefit and favor short-termist thinking for a single generation, which the boomers have done.
- 2:35:13The boomers have destroyed Western civilization.
- 2:35:15But in that model, in that example, describe the process of it going back to long-term thinking.
- 2:35:22What happens then?
- 2:35:24Because why doesn't capital just keep flowing to the quick, make your quick bucks, people?
- 2:35:29It does.
- 2:35:29But eventually given a long enough time horizon if that firm that investing can hold on can stay profitable can stay alive long enough for the things they investing in to mature they have created entirely new business categories entirely new products and services which the existing firms then have to try to scramble to try to keep up with They can do anywhere near as well because they haven put the time and effort into actually developing it So over a long enough time horizon this ends up
- 2:35:55ideally working out. The problem is when you have enough of a broken system where those firms can't
- 2:36:01survive long enough, then it's like a circuit breaker almost. Something has to give. There has
- 2:36:06to be a crash, a recession, a reset. Something has to give to reset the conditions such that
- 2:36:12those long-termists can work out because you're right. The short-termists can keep winning until
- 2:36:17there's no more road down which they can kick the can, right? And we're rushing toward a bunch of
- 2:36:22cliffs that are just can kicking. The federal deficit in debt and social security and federal
- 2:36:29spending writ large in the United States, the fertility crisis at a civilizational scale.
- 2:36:33There's a bunch of these inflation and the devaluation of the monetary base to – the Bitcoiners are totally right about that.
- 2:36:39There's a bunch of these can-kicking things we are just doing as a civilization.
- 2:36:43Eventually, the road runs to a cliff, and you kick the can off the cliff, and then you can't get it.
- 2:36:47And you accidentally fall off the cliff because you weren't paying attention.
- 2:36:50And you kept saying, oh, well, I could kick it again.
- 2:36:52I see that there's a cliff up there somewhere.
- 2:36:54I get it, but we're not at it yet.
- 2:36:55I can kick it again.
- 2:36:55I can kick it again.
- 2:36:56And eventually, you kick a can off the cliff, and then there's some sort of a reset.
- 2:37:01And it's not clear to me, to be totally honest.
- 2:37:04I don't know what this looks like.
- 2:37:05I don't have specifics beyond this.
- 2:37:07But it seems to me at a high level, like civilizational cyclical pattern scale, you eventually end up with such decadence that short-term thinking reigns until the can gets kicked off the cliff.
- 2:37:17There's some sort of reset of some sort.
- 2:37:19I don't know what that looks like.
- 2:37:20And then the long-termists who were the whole time building toward a longer-term civilization, a more prosperous in the broad human flourishing sense civilization, are the only ones left, more or less.
- 2:37:32Because they're the only ones that have, in our current case, the only ones that have reproduced.
- 2:37:35There must be an AI component of what you're talking about.
- 2:37:39Because if anything, AI turbocharges the dopamine people.
- 2:37:48To the nth degree.
- 2:37:48Right.
- 2:37:49so you say well what's the what's what does it look like wouldn't it be that isn't that the
- 2:37:56hockey stick isn't that the the part of that end of the civilization where it just goes absolutely
- 2:38:01rocket ship and it also because it it it it it multiplies everything including job losses including
- 2:38:11like you say sorting out basically sorting out the dopaminers and the long-termers that's what
- 2:38:17it's doing if that if that thought is really true and there's there's a there's winners and losers
- 2:38:22there and it's pretty obvious who wins and who loses well we already see it happening so my wife
- 2:38:27is a pediatric occupational therapist what that means is she works with developmentally delayed
- 2:38:31disabled you know kids that are having trouble with a variety of things you know like 15 and
- 2:38:36younger ish or the usual age range is like from 2 to 10 12 right in that range and this is already
- 2:38:42starting to occur because she gets kids regularly that have terrible like social interactions,
- 2:38:50terrible behaviors. They have terrible emotional regulation, bad posture, bad physical coordination
- 2:38:57because they're iPad kids. iPad kids are effectively this dead end. They are – the
- 2:39:03way to make this visceral for people is go watch the movie WALL-E. The people in the beginning?
- 2:39:07WALL-E.
- 2:39:07WALL-E. So in WALL-E, the earth is covered in trash and you're following this robot,
- 2:39:10cleaning it up, whatever. Well, the reason the Earth is covered in trash is because humans are
- 2:39:14in space and spaceships all like plugged into the matrix. They're all like morbidly obese,
- 2:39:17you know, screen here and there's a helmet, you know, they're plugged in completely.
- 2:39:21That's the iPad kits. It's just we're doing it with iPads instead of like full body immersion
- 2:39:25kits, but you'll get, well, we'll get there. The AI advances enable places like Meta, which is one
- 2:39:31of the most evil companies on Earth, to get massively better at optimizing the capital A
- 2:39:37algorithm to get people more hooked on reels, more hooked on Facebook, more hooked on Instagram,
- 2:39:41to keep them there more, to destroy attention spans more, to destroy productive capacity more
- 2:39:45for their own short-term selfish gain and profit until that accelerates so hard that I don't know
- 2:39:53what happens. So this is the AI makes this very hard to reason about because the outcomes are so
- 2:39:59absurdly divergent. You have the Yudkowskian like gray goo scenario where recursive super
- 2:40:05intelligence just converts everything into what they call computronium as a kind of hand wavy
- 2:40:10thing. It was basically like more computers to fund more, to power more inference. And they
- 2:40:13start tiling the solar system in the galaxy and computronium effectively. Like that is a
- 2:40:17within the realm, the way I evaluate these things is like, does it violate the laws of physics?
- 2:40:21No. Then it is at least in principle, not saying it's plausible, but it's in principle possible.
- 2:40:26That's a possible outcome. Wild human flourishing, solving all disease and aging and biological
- 2:40:32suffering, wireheading, as they call it, right? Which is like just directly plugging a computer
- 2:40:36into your brain to just have the neural link, the whatever, turn off the suffering receptors and
- 2:40:42turn up the dopamine receptors to the max, like within the brain itself in an inescapable way,
- 2:40:47where everybody ends up plugged into the matrix, except for like a select few who are running set
- 2:40:51and own and control said matrix. That doesn't violate the laws of physics. That's a plausible
- 2:40:55outcome. You have the simulation hypothesis scenario where eventually we don't see advanced
- 2:40:59civilizations. It's a solution to the Fermi paradox because everyone's plugged into simulation. So
- 2:41:04when societies get advanced enough, they just go to simulation land and they stop expanding in the
- 2:41:08physical universe because they're just in simulations. And the simulation is infinitely
- 2:41:11programmable to be pleasurable. So why not just stay there? Like there's so many divergent scenarios
- 2:41:15here. Everything I am saying now has a giant asterisk, which is like, oh, and AI. I don't,
- 2:41:21there's so many outcomes that are just in principle possible.
- 2:41:24what um what's your most likely let me let me tell you how i how i see this playing out
- 2:41:32we end up with a two-tier society because we have to and i've never heard of the sort before but
- 2:41:39that basically backs up this the two the two-tier society is basically what you described in the
- 2:41:45sort and within that within that world there will be people who choose to plug in to the matrix
- 2:41:53literally who are the ones who basically lost in the sort and the people that win in the sort will
- 2:41:59either be the evil authoritarians or there'll be people who know how to opt out intelligently
- 2:42:07and understand that dopamine is an important element of the human condition but there are
- 2:42:15much deeper and more meaningful ways to live. Because even technology and this pursuit of
- 2:42:23constantly iterating over and over again, when you actually meditate on it is a bizarre thing
- 2:42:30in itself. Because how less meaningful is my life running a farm in the Himalayas than being at the
- 2:42:41forefront of AI. I mean, really, really, and honestly, and seriously, in the end, what's more
- 2:42:48important? Who has the more meaningful life? Who has the better life? And it could be that a large
- 2:42:55section of those, if the sort happens, the winners in the sort, a large section of them,
- 2:43:02are just living quite a simple life, actually. Because especially in a world of abundance,
- 2:43:07if if the if the ai abundance thing does occur most of the things that we take for granted as
- 2:43:14having meaning now in the physical world will become meaningless because you'll be able to
- 2:43:18i mean you know this is a wild prediction but 3d print anything you want at any moment you have
- 2:43:26the jet sentence like little whatever yeah i mean it's not hard to believe that at least one day
- 2:43:31that's going to exist so so as a result what what's the no one there's nothing to there's nothing to
- 2:43:38persevere there you just press a button so what do you persevere on i think we talked about this
- 2:43:43before i think it what is that it's it's the human condition the spiritual world it's the
- 2:43:51it's the that area of the subconscious that you don't know what you go on a journey of discovery
- 2:43:56to try and find out who you are, what you are.
- 2:43:59Maybe you discover it is a simulation
- 2:44:01and wow, that would be really disappointing.
- 2:44:04Or would it?
- 2:44:05I don't know.
- 2:44:06Yeah, that's a whole fun rabbit hole in itself.
- 2:44:09No, I think what you're describing is effectively
- 2:44:12my kind of most likely outcome as well,
- 2:44:17which is the majority of people call it,
- 2:44:19I'm making up a random number,
- 2:44:20but 75, 80% of people do like
- 2:44:22whatever the version of plugging in is.
- 2:44:23It's the bread and circuses.
- 2:44:24It is the –
- 2:44:26Which is already true now.
- 2:44:27Well, the reason I'm saying this is because it's not an – it's a very short extrapolation from the current circumstances more or less.
- 2:44:33It is like the TikTokification of the rest of reality more or less.
- 2:44:37And then I think after that, you sort into a variety of camps.
- 2:44:41The two biggest are probably like the more authoritarian, like the people that run the AI companies that own the data centers that are creating the thing that the people and control the thing that people are plugging into.
- 2:44:50and you have the group of conscientious objectors that for principled clear reasons opt out.
- 2:44:56And I think there's going to be, again, it's like a 10 to 20% of the population that ends up in the
- 2:45:01latter camps of opting out in some way, shape or form because it requires enough willpower and low
- 2:45:07time preference to not get sucked in by the siren song of plugging in in and of itself.
- 2:45:12I think what's really illustrative to like see what this could look like is look at
- 2:45:15guys that exit from a startup, they make a bunch of money. They have a windfall of some sort from
- 2:45:21really hard work, especially because it's that person out, not lottery winners, but people that
- 2:45:24worked really hard and made a bunch of money. What do they usually do afterward? They throw
- 2:45:29themselves into cultural or spiritual pursuits. They get really into painting, gardening. They go
- 2:45:35start foundations and philanthropies that are in the more cultural, spiritual side of existence.
- 2:45:41There's a great meme that circulates on X every now and then where it's this guy's LinkedIn and
- 2:45:44It was like hedge fund manager for like 30 years at some big Wall Street hedge fund.
- 2:45:48And then the next job on there was a local gardener.
- 2:45:51That's what he's been doing for the last like 10 years.
- 2:45:53He just runs – he has more money than God.
- 2:45:55He just like has a little garden that he keeps for the community and feeds his neighbors with.
- 2:46:00And it's like a very simple, deep, fulfilling life actually.
- 2:46:04So it's a peek into what a post-scarcity world could look like by looking at the few humans who are currently right now post-scarcity.
- 2:46:09What are they doing?
- 2:46:10What are they up to?
- 2:46:11They typically find new missions to go hunt.
- 2:46:13Like Jeff Bezos started Blue Origin, right?
- 2:46:16That's what I'd say.
- 2:46:18The few founders I know personally who've done that tried to do nothing or tried to do something simple, but always ended up going back and founding something else.
- 2:46:29I mean, like, you know, I don't know anyone who's really stopped.
- 2:46:35It's because it gives them purpose and meaning to go, like, create value, to do hard things, to go try to do those things.
- 2:46:39But gardening can be hard if you decide to create something specifically in the garden I know it not different I think a lot of what drives that is the earning of the money as well actually I think getting big paychecks is a massive incentive
- 2:47:00even if you don't need the money.
- 2:47:02It's still-
- 2:47:03Because it's a social status signifier.
- 2:47:04So that's my other prediction for the outcome of this,
- 2:47:08which is material scarcity is like more or less solved,
- 2:47:11which by the way, if you live in like the United States
- 2:47:13or, you know, the more prosperous countries,
- 2:47:15that is already the case.
- 2:47:17Like if you go interview,
- 2:47:18there's great YouTube channels to do that. Just look up homeless man interview. Like literally
- 2:47:21just look that up on YouTube. And you'll see time and again, people just man on the street
- 2:47:24interviewing homeless people. And they are either like genuinely like psychotic, like literally
- 2:47:29their mind is broken in a fundamental way, which is, I feel deep empathy and sorrow for those
- 2:47:33people, but you know, understand why they're there. There's another big chunk who are just
- 2:47:36like, this is awesome. I don't have any responsibilities. I can just go wherever I
- 2:47:39want. I don't have to do anything. People give me free food. I just kind of mess around all day.
- 2:47:43They might not even be on drugs. They're just like, yeah, I just kind of, a lot of them are
- 2:47:45on drugs. They're dopamaxing. But the ones that aren't even are just, yeah, this is a great life.
- 2:47:50I'm actually kind of deeply enjoying this. And they typically end up there for a reason. My point
- 2:47:54there is just like we already live kind of in post-carcity. And so far, it's like you're not
- 2:47:58going to literally starve on the street. It is hard to work yourself into a circumstance where
- 2:48:02that happens in modern America, at least. So we get a glimpse of kind of what this can look like
- 2:48:07now. So I think what might end up happening in that situation is everything just becomes about
- 2:48:12social signaling and social status and hierarchy because material scarcity, social status is
- 2:48:18inherently and absolutely relative, right? So you can have a society on the brink of collapse.
- 2:48:24It's like an agrarian hunter-gatherer type society that barely gets enough food to feed
- 2:48:28their population month over month. And post-scarcity AI utopia, the one invariant that
- 2:48:34is identical across both of them is social hierarchy because it is a relative rank order.
- 2:48:38So by definition, it is always a zero-sum scarce resource game. It's always zero-sum.
- 2:48:43So there's also a world where what happens is the bread and circuses, the dopamaxing, ends up just becoming effectively increasingly absurd social status games and social signaling to try to climb relative social hierarchies, which also obviously is not like spiritually fulfilling or deeply.
- 2:49:00it's not a world of human flourishing at large. But again, there's a subset of people that right
- 2:49:05now, and therefore then as well, will be able to opt out of those and just live a fulfilling life
- 2:49:08and social status be damned. And that subset of people will end up shaping and sculpting whatever
- 2:49:15world comes next because the rest of them are not participating in the shaping of what comes next
- 2:49:18by definition because they're focused on short-term social status games.
- 2:49:21What would social status be in a post-scarcity world though?
- 2:49:25This is where things get super weird because I don't have any idea. I don't think anybody does.
- 2:49:28It could be things like, I mean, it's just how many other human beings are at your command.
- 2:49:35That's kind of what it is.
- 2:49:36Or I know in this day and age, it's things like when you get so rich, you just want a hospital named after you, don't you?
- 2:49:43Or a library or something.
- 2:49:44Right, exactly.
- 2:49:45Because that would probably still stay.
- 2:49:50I mean, you're not going to be able to 3D print hospitals, mind you.
- 2:49:52We probably won't be going to hospital, will we?
- 2:49:55You don't have to.
- 2:49:55There'll be little nanobots flying around us just saying, oh, look, you need a little bit of work.
- 2:50:00Just stand still a minute.
- 2:50:01There you go.
- 2:50:02Exactly.
- 2:50:02You're good.
- 2:50:03Exactly.
- 2:50:04No, it gets really weird when you try to think about it like really fast because humans, if nothing else, are really good at making up status games.
- 2:50:13Like we're very good at that.
- 2:50:14Right now we see a preview of this with the balkanization of the internet.
- 2:50:16So everyone's in their nice little internet social hierarchy and status clique.
- 2:50:19So I'll just use one that I know of personally as a silly example of this.
- 2:50:24That game I talked about I played earlier, Rocket League, has a deep online internet subculture with deep social status and social hierarchy that you would only know if you're deep into the Rocket League community.
- 2:50:35But there's intense status competition there.
- 2:50:38Literally, these guys, the kids that are the professionals at this – there's a pro league, obviously.
- 2:50:42The kids that are professionals at this will – they do like things called show matches, which are not for money.
- 2:50:47They're not for anything.
- 2:50:48They just do them.
- 2:50:49People watch them on YouTube.
- 2:50:50They'll get hundreds of thousands of views just to prove who's better at 1v1, right?
- 2:50:54It's all just status games all the way down.
- 2:50:56You have this in TikTok, Instagram, social media are just pure, unadulterated status
- 2:51:00games very directly.
- 2:51:01There's a numerical measure, your followers, your likes of your social status there directly.
- 2:51:06You see this in every little internet subculture.
- 2:51:08You get into the knitting and crocheting community.
- 2:51:11There's going to be a social status hierarchy there where there's one person in the community
- 2:51:14that everybody kind of follows and listens to that comes up with the new ideas.
- 2:51:17So you end up with a like fractal kind of disruption of society in intensification of the trends we already see of this balkanization of society into niche subcultures with their own inherent status hierarchies.
- 2:51:31So what happens is there's not like a global unified status hierarchy, which we currently have, which is proxied by how much money you make effectively across all of civilization.
- 2:51:39And it's instead – that matters a lot less.
- 2:51:42So it connotes less social status.
- 2:51:43So these deep niche sub-communities and social hierarchies experience much more intense and rapid social hierarchy competition.
- 2:51:53If you look at the speedrunning world, it's another great example of this.
- 2:51:56There's a whole sub-community internet subculture of speedrunning old games.
- 2:51:59How fast can you complete them?
- 2:52:00There was a huge breakthrough in Super Mario 64 from the Nintendo 64 recently.
- 2:52:07A player, so they've calculated out the exact inputs, like the exact frame level, frame by frame inputs you need to get a perfect speed run.
- 2:52:16And a human recently, obviously AIs have just beaten this fully, but they don't care because it's critical.
- 2:52:22AI solved it years ago, but it doesn't matter.
- 2:52:24Like a frame perfect run.
- 2:52:26No one cares because there's no status connoted with that whatsoever.
- 2:52:28No one cares at all.
- 2:52:30But a human recently made waves.
- 2:52:31The video has millions of views right now.
- 2:52:33It's like five or six million.
- 2:52:33A dude got a run that went from the previous PR of the dude only missed five frame inputs down to three. And now they're debating if it's even physically humanly possible with human reaction times to get below three.
- 2:52:45absurd silly example to make my point that we are insanely good at inventing social status games and
- 2:52:54social hierarchies so that world just looks so weird like who would have predicted x ante that
- 2:52:59a video of a speedrunner getting a uh only missing three frames on a super mario 64 speedrun would
- 2:53:05have millions of views because that subculture happens to be that big that's crazy i saw i saw
- 2:53:10that that there was a famous meme going around recently and it was a guy doing guitar hero
- 2:53:14in the most insane way.
- 2:53:17I think the hardest level
- 2:53:18and he got 100%
- 2:53:19and underneath it,
- 2:53:20someone wrote,
- 2:53:21you know,
- 2:53:22200 years ago,
- 2:53:23this guy would have been on a ship
- 2:53:24traveling around the world
- 2:53:26making maps of the world,
- 2:53:28you know.
- 2:53:29The way I found out
- 2:53:29about this Mario thing
- 2:53:30was a guy quote tweeting that tweet.
- 2:53:32Oh, right.
- 2:53:33He was like,
- 2:53:33here's an example
- 2:53:34and it was that video.
- 2:53:35That was how I found out about this
- 2:53:36was that exact tweet.
- 2:53:39I just wondered
- 2:53:40whether a social hierarchy
- 2:53:42might arise
- 2:53:43around self-actualization because because it in a way it can't you know real self-actualization
- 2:53:50i was trying to think you know say take take go back to this idea of the sort and the the so-called
- 2:53:57winners from the sort will basically hopefully have transcended social hierarchy um because
- 2:54:03what you're pursuing at that point if it's something like self-actualization doesn't work
- 2:54:09if you're pursuing a social hierarchy.
- 2:54:11Like the whole point is it's the opposite of that.
- 2:54:13It's the instrumental versus whole of self commitment
- 2:54:16I was talking about earlier.
- 2:54:17It's another version of that.
- 2:54:18If you are instrumentally pursuing self-actualization,
- 2:54:20you will by definition be unable to reach it.
- 2:54:22Right, okay.
- 2:54:23So that might be,
- 2:54:24I mean, that might be one of the saving graces of society.
- 2:54:29I don't know.
- 2:54:30Like the more,
- 2:54:30because you come to realize that, hopefully,
- 2:54:34but you do come to realize it.
- 2:54:35So if you end up in the correct side of the sort, you should realise that.
- 2:54:43And there should be a pretty good life to be had amongst other people who worked that one out as well.
- 2:54:50And I don't know, arguably, you could look back at previous civilisations and see, you know,
- 2:54:56like I was just thinking of the pyramids, these strange, weird triangles.
- 2:55:01What's that all about?
- 2:55:02But someone built them and someone built them in a very specific way for a very specific reason that seems a bit strange now.
- 2:55:11But they did it.
- 2:55:13And that's the kind of thing that I think people, whatever pyramids are for, I don't know.
- 2:55:20I just know they're weird.
- 2:55:21I know they're unusual.
- 2:55:23I know they're very specifically built and they're ginormous.
- 2:55:28And someone really decided that they wanted to build those things.
- 2:55:32And we have no idea what they're supposed to do.
- 2:55:34They're not tombs.
- 2:55:35I think that's nonsense.
- 2:55:37But in that world of the sort, when you get, hopefully, you also get a lot of the keenest minds and the brightest hearts in that echelon or that sector.
- 2:55:50That's when people start pursuing those kind of things again.
- 2:55:54And they'll also be very long term.
- 2:55:57That's exactly right.
- 2:55:58I think – I guess the only upside of the sort because it has many, many downsides is that it eventually bottoms out in this world kind of similar to this, which I think comes full circle to a point you kind of alluded to earlier that I thought a lot about recently because it's one of those things – I'll pose the question.
- 2:56:16And a lot of people reflexively have an answer, but then if you actually think about it, there's – the answer is not clear to me at all, which is like imagine two people.
- 2:56:24One guy is like a finance bro, lives in Manhattan.
- 2:56:28He did an investment banking internship at Goldman, and now he's in some private equity firm, and he works 100 hours a week, and he makes $10 million, $20 million a year, whatever, and seems like he's totally crushing it by like all objective metrics.
- 2:56:42And then you have the guy who has a – let's say he has a farm in Mississippi.
- 2:56:47I'm actually like describing an uncle of mine.
- 2:56:49He has a property in Mississippi.
- 2:56:52He has a chicken farm.
- 2:56:53He has a landscaping business.
- 2:56:54He has a little grave digging business, all like very local community oriented.
- 2:56:57They're all kind of small in and of themselves.
- 2:56:59Makes enough money to like keep his family comfortable and go on like one vacation every
- 2:57:03year-ish, give or take for five days or something like that.
- 2:57:08And is far more – has far more equanimity than I do, far more tranquility than that
- 2:57:17finance private equity guy ever will.
- 2:57:20And so if you pose the question then superficially, which of these guys has a better life?
- 2:57:26Which one is succeeding more?
- 2:57:27Well, the gut answer is obviously the private equity guy is crushing it.
- 2:57:29He could do anything.
- 2:57:30He has millions and millions of dollars.
- 2:57:32But if you then think about, okay, but we all have around 2 billion heartbeats in our life.
- 2:57:37That's the upper limit on basically all animals, 2 billion heartbeats.
- 2:57:41If that's the case, then how are they – the subjective experience, heartbeat to heartbeat,
- 2:57:46in any random snapshot moment of the day of my Uncle Joey and this guy, this private equity guy
- 2:57:52I'm making up, which one has a better subjective, which one has better qualia, which one has a
- 2:57:57better subjective experience heartbeat to heartbeat on any given snapshot of a day? If you ask
- 2:58:02That way, the answer is not at all clear.
- 2:58:04And in fact, I would argue my uncle has a better subjective experience than Mr. Worth
- 2:58:10to quarter billion dollar private equity managing partner.
- 2:58:13That guy has a terrible, terrible, probably the ones I've met, terrible, stressful, anxious
- 2:58:20day-to-day lived experience because they're at the very top of a status hierarchy, which
- 2:58:23means they're precarious.
- 2:58:24One slip up, there are hordes of people chasing them, trying to get that status hierarchy spot
- 2:58:28because again, it's zero sum by definition.
- 2:58:30my uncle has opted out of that completely he doesn't care he just loves the life he lives
- 2:58:36he tends to his chickens and his he landscapes the local you know public spaces and the local
- 2:58:42community college and there's like two graveyards in the town and he digs all the graves for them
- 2:58:47and he's a pillar of the community and everyone loves him i think he's like a county board of
- 2:58:51supervisor superintendent now he loves his life he is the one of the most tranquil like unflappable
- 2:58:58calm, seemingly joyful people that I've met just day-to-day existence. Compare that to
- 2:59:05the stress ball of life that is Mr. Private Equity. So I don't know which one of those
- 2:59:10ends up leading a better life, but the fact that that is ambiguous, I think hints at
- 2:59:15what a post-scarcity future could or should at least look like where the vast majority of people
- 2:59:20are stuck in status hierarchy, anxious, social anxiety like that guy. And then some of us are
- 2:59:26lucky enough to realize that whole rat race is stupid self-actualize and opt out entirely it's
- 2:59:30it's meaning again though and in the case of the meaning i think the most the most universal form
- 2:59:39of meaning is service to others and i think if you look at the guy the hedge fund guy is servicing
- 2:59:48himself your uncle is probably service is it is in service to other people and i think that's where
- 2:59:56meaning comes from every month he digs a few graves for free right because it's it's just
- 3:00:00indigent people in the community we know personally but that's a life hack realizing that's the same
- 3:00:05as people who start to realize that taking on responsibility is liberating because it seems
- 3:00:11like an oxymoron it's the same service being in service to others is is amazing it's it's the
- 3:00:18opposite of what the hedge fund guy thinks it's like i'm not going to do shit for free i want what's
- 3:00:22mine kind of thing and it's the opposite and this comes back to my theory that consciousness is an
- 3:00:28outward flow love is not something you receive it's something you give the feeling of love is is
- 3:00:35is the feeling of love coming out from you it's not that most i think most a lot of people frame
- 3:00:41it as when you're in love with someone, I'm getting love from that person. It's the opposite.
- 3:00:46You're feeling your own love that's projecting outwards. Totally. Anyone that's, I think,
- 3:00:52married and has a good marriage knows this is, can immediately intuit this is true. Because if
- 3:00:57you ask a person in a true, wholesome marriage, what is going to bring you more joy? That you
- 3:01:04bought a new car or that you bought your wife a new car? Every married man that's listening knows
- 3:01:10the answer to that immediately. The look on her face when you did the thing, whatever it is,
- 3:01:14the act of service, the thing for her, which is an outpouring of love from you to her,
- 3:01:21you can see in your mind what that – her reaction, what that's going to be like,
- 3:01:23seeing the just unadulterated joy. That is why you do that. That is why you work super hard and do –
- 3:01:29that's what literally at a very basic level, what motivates me to work as hard as I do,
- 3:01:33is literally knowing, okay, doing this and gathering the resources needed to be like
- 3:01:38financially comfortable will enable my wife and my hopefully, you know, knock on wood, soon to be
- 3:01:43future children to live joyous, stress-free, like wonderful lives. And sure, the whole time I have
- 3:01:53to keep working, I have to keep stressing or doing whatever, you know, working hard, grinding,
- 3:01:57but it is not only worth it, but actively a joyful act because I'm doing it in service of others.
- 3:02:02I mean, what did, again, coming back full circle to what we said earlier, right? Like what did
- 3:02:06Christ say in Matthew, love thy neighbor as thyself. Most people like superficially read
- 3:02:11that as some sort of like, oh, be nice. It's this, what we're talking about right here. That's what
- 3:02:16that verse means in context. Not a generic, oh, be nice to everyone. It is this acts of service of
- 3:02:21the outpouring of love to others that results in full self-actualization and realization
- 3:02:27as a fully realized human. You see the same thread of thought in all of the Lindy, meaning like
- 3:02:33multi-millennia old religions, right?
- 3:02:37Ancient Judaism and traditional Judaism is very like community outward service oriented.
- 3:02:43Buddhism is very like community outward service oriented.
- 3:02:47They all settle on this like rough approximation of the golden rule.
- 3:02:51And it's because the golden rule is not some banality.
- 3:02:53It is this deep truth of the true source of love, of purpose, of meaning, and therefore
- 3:02:58of living a flourishing existence.
- 3:03:00i'm glad we've ended up at the golden rule anyway it was only a matter of time we get there
- 3:03:06eventually we have to get there eventually every conversation about spiritual matters
- 3:03:11terminates at the golden rule no but when you were saying that i was thinking that
- 3:03:15you don't even have to get woo about that in the sense that if you are in service to someone
- 3:03:26It's not a self-sacrifice even.
- 3:03:30It's because you're feeling that thing.
- 3:03:32You're the one feeling it.
- 3:03:33It's selfish.
- 3:03:34Exactly.
- 3:03:35It's selfish in the Ayn Randian kind of good way, I think.
- 3:03:39She has a great essay about this.
- 3:03:40Does she?
- 3:03:41Mm-hmm.
- 3:03:44It's in the Ayn Rand Anthology, which is this collection of works that Ayn Rand Institute has published.
- 3:03:48It's one of the essays in there.
- 3:03:48I can't remember which one.
- 3:03:50But she talks about love.
- 3:03:51So it's an essay about the selfishness of love.
- 3:03:53And she makes this exact same point that like actually people that love the deepest and do the most for those they love are the most selfish people in her definition of selfish, which has a positive valence obviously.
- 3:04:03Because like what you're doing it ultimately because it makes your life fulfilling and worth living.
- 3:04:07Like yes, the mechanistic action is you're making someone else's life better.
- 3:04:11But you're ultimately doing it because it feels great for you.
- 3:04:13And also, as an aside, as you were saying, you don't even have to be woo or spiritual about this.
- 3:04:19I think it speaks to how broken culture is generally that you and I both through this whole conversation have had to caveat like every other line with something like that.
- 3:04:26Isn't that correct?
- 3:04:26We can't just like have a direct conversation for people listening.
- 3:04:29We have to caveat or else they won't even take it seriously with, oh, we're not being like super woo here.
- 3:04:34There's a rational explanation for this.
- 3:04:35You can't even just have the direct conversation.
- 3:04:36I noticed that as you were speaking, I was like, man, he and I both have caveated so many things we've said with.
- 3:04:43Well, there's a rational explanation for this.
- 3:04:45Yeah, but I think that's a good thing because I think you never know who's listening.
- 3:04:50That's true.
- 3:04:51And I want people to understand what I understand.
- 3:04:55I feel fortunate to live the way I do.
- 3:05:02And I know that it's just a function of how I operate in the world.
- 3:05:06You know, it's not some, it's not woo-woo.
- 3:05:11Well, it kind of is.
- 3:05:11but you but you know and look this is you know i was gonna say anyone can do anything i don't
- 3:05:18think that's true actually not anyone can do anything but i think in this in the realm of
- 3:05:22what you call self-actualization i think anyone can do anything correct you know you might not
- 3:05:28you might not be able to be a super ai understander like you are i'm not i'm never going to be a lead
- 3:05:34singer for a popular band you never know you've never heard me say don't say it out loud you've
- 3:05:41never heard me sing, but anyone can self-actualize. That's another key point of why this is the
- 3:05:45correct, like kind of vector trajectory to be on. I think just for anyone is because like the
- 3:05:51material outcomes you can achieve are delimited by like biological material, circumstantial
- 3:05:56reality. Like there are, I think of human performance in any given field as like any
- 3:06:01given person has like a band in which they can perform. And if you train really hard to try your
- 3:06:05hardest, you can get to your full potential. And if you never do it ever, then maybe you never get
- 3:06:09even close to your full potential, but then everyone has different bands effectively.
- 3:06:13Like I, I, I could have started training swimming from the day I was born and I'm never going
- 3:06:18to beat Michael Phelps because he is like weirdly just a genetic anomaly optimized for
- 3:06:22swimming with his limb proportions, et cetera.
- 3:06:23So like there's a, there's a, you know, a spectrum on which everyone can, can perform.
- 3:06:29And I think self-actualization basically just means becoming the best version of you becoming
- 3:06:36a holistic unified self, they're limited by those bands. So there's like a, just what's physically
- 3:06:41possible for you. But within that, the, the, no matter how narrow or wide your band is or where
- 3:06:47your band is, anyone can self-actualize. Self-actualization doesn't actually have any
- 3:06:52impact one way or the other on your like material outcomes, unless that's part of your self-actualization
- 3:06:57journey. And, and, you know, it's going to be wildly different for each individual person,
- 3:07:01but I think there's a, there's just a spectrum in which everyone can perform, but anyone can
- 3:07:05self-actualized that's the key point here but going back to what you were saying about social
- 3:07:10hierarchies there is a version of that it's not a but it doesn't carry a negative connotation for
- 3:07:16me in this realm in as much as when i see someone who's really self-actualized i i want to be like
- 3:07:24them you know they are i like i acknowledge wow this is an inspirational person and actually it's
- 3:07:31but they're not trying to claw their way to the top of the social hierarchy,
- 3:07:35but you can still exist within that hierarchy and acknowledge that hierarchy.
- 3:07:40And I'm going back to the sort again, you know, like this, what does that,
- 3:07:45what does that world look like when you end up in it?
- 3:07:49Because I hope I do, because according to the alternatives,
- 3:07:53I definitely want to end up in it. Yeah. Like, um,
- 3:07:56so I expect to end up in there.
- 3:07:58I expect my kids and my family to end up in that world.
- 3:08:01as well and i'd say i've never i've never heard it described as the sort but i really do like that
- 3:08:08anecdote i'm already preparing for it and have been for a number of years already which is with
- 3:08:15my kids is making sure they understand that um they need to position themselves very well in the
- 3:08:23in the next 10 20 years and they need to be very flexible in in where they end up because um what
- 3:08:32you don't want to be in these next 20 years is tied down to something that you you can't exit from
- 3:08:40that's exactly right and i i think there's a
- 3:08:44oh let's return to thought uh i had a directly uh i had a response to that son of a gun
- 3:08:50um it was on the self-actualization point give me a second oh my gosh i lost it completely
- 3:08:57i hate it when that happens then the harder you try to remember it the further away it gets start
- 3:09:02talking again yeah oh i remember uh so when it comes to this like hierarchy within self you totally right i run into people occasionally who I like wow You can again frequency vibe
- 3:09:15You pick up on it, especially if you're down this journey at all.
- 3:09:19You pick up on it immediately.
- 3:09:20Like, oh, that guy is further along this kind of actualization and kind of integration journey than I am.
- 3:09:27But that does not mean that you can or should.
- 3:09:32And in fact, if proper self-actualization, I would argue, like almost precludes this result by default because it would mean you're not self-actualized.
- 3:09:39You don't attach meaning or emotional valence to the fact that that guy is like more spiritually integrated, more holistically integrated than you are.
- 3:09:48You see it more as, ah, there's a – I can go further.
- 3:09:52Like there's a further place to get to.
- 3:09:54So there's this mental framework I like to use for almost everything that can help you process, work through things, which is like the story.
- 3:10:01So what happened, just the raw facts of what happened and the story you tell yourself about
- 3:10:07what happened. Okay. So let's take an example of, you know, I'll give a real one from my own life.
- 3:10:13Okay. When I was in college, this like long-term girlfriend, like high school sweetheart broke up
- 3:10:17with me unexpectedly. Okay. That's the fact of what happened. A relationship ended. A relationship
- 3:10:22ended. That's it. The story you tell yourself about what happened completely determines what
- 3:10:27happens next. One story you could tell is, oh, this broke me forever and I'll never get over
- 3:10:34this. And this means I'm fundamentally unworthy of future love, blah, blah, blah. Or it could mean
- 3:10:40to you, and therefore I'm very angry about this and I hate all members of the opposite gender and
- 3:10:46I'll never fully commit or love again fully because I don't ever want to get hurt again.
- 3:10:50or you can create a different story. You could say, okay, that happened and it was a learning
- 3:10:57experience for me. I did not perform well in that relationship. I've learned a lot of things that
- 3:11:01will make me a better partner in future relationships. Move on. My point is all of those
- 3:11:06are stories that you can tell yourself about what happens that creates the emotional and
- 3:11:11psychological valence of the thing, but they are all things you can choose because the underlying
- 3:11:17ground reality truth is a relationship ended. That's it. The rest of the implications that you,
- 3:11:23and we all attach automatically and unthinkingly, are all just that. They are stories you are
- 3:11:29telling yourself about a fact that occurred. The fact in and of itself carries no emotional,
- 3:11:36psychological, or spiritual weight whatsoever. It is just a thing that happened. The same way
- 3:11:41if you drop a rock and it rolls down the hill, the rock rolled down the hill. A relationship ended.
- 3:11:44I mean, that's the basis of Viktor Frankl's search for meaning, isn't it? Real freedom exists in the way you respond to reality. That's your freedom. That's your ultimate freedom. And I agree with that, I think.
- 3:12:02And the fully self-actualized person just – it can consciously orient how they respond. It's – the core lesson of stoicism is exactly the same thing. There may be material things in reality that you cannot change. But A, the stoics I think properly construed would argue there's a lot more that you can impact than you think.
- 3:12:18But assume a bad thing happened and you are saying it's bad.
- 3:12:22There's a story you're telling yourself about it.
- 3:12:24And you did everything you could to prevent it and counteract it and whatever.
- 3:12:28Well, then the Stoic would argue a very similar thing.
- 3:12:30Well, then the story you need to tell yourself about that is not, oh, woe is me, this hurt, whatever.
- 3:12:34It is simply accepting that it happened, looking at what is now under your control and moving forward.
- 3:12:39Because guess what?
- 3:12:40The past does not exist.
- 3:12:42It does not exist.
- 3:12:43The only thing that exists is the present moment.
- 3:12:45set aside like special relativity uncertainty about what present actually means across large
- 3:12:49distances because of the speed of light. Only the present exists. The past is gone. It does not
- 3:12:57exist anymore. And the more time you spend worrying and fretting and ruminating on it,
- 3:13:02is just you wasting the present reality on a fiction that no longer exists. There is just
- 3:13:07the fact of physical reality of an occurrence and then the next moment.
- 3:13:11Which is also the flow you're talking about. The exact definition of it.
- 3:13:15because you don't need to analyze what's happening because actually analyzing is taking you out of
- 3:13:20the moment it literally is it's it's putting you in a secondary moment looking back actually you're
- 3:13:27looking back at what's actually happening and yeah i mean that's that's the long i mean that's one of
- 3:13:33the most important journeys in i've never called it self-actualization but actually it sounds
- 3:13:38it conjures it up quite well the most important journey is to exist in the moment
- 3:13:45yeah is to learn to to exist in the moment and and to to trust the process because you you have
- 3:13:56to stop analyzing what's going on basically there are times to use your brain you need to add things
- 3:14:01up and you need to interact with you need to fix two wires together on my mic and you need to do
- 3:14:06all that shit but on the whole the rest of it is not necessary to use the kind of calculating part
- 3:14:13of your brain actually let's go back we've got to wrap it up soon mate that's well over three hours
- 3:14:20dude awesome it feels like i've been here for an hour at tops maybe have you ever taken um
- 3:14:26loosener jets of course right this feels like the morning after afterglow it feels kind of nice
- 3:14:31you're kind of tired totally um we've got to go back to ai because that's where we started and
- 3:14:39the moment and consciousness and all this kind of stuff what's what do we think what's our latest
- 3:14:44what do we think ai is like what in that in that perspective like is it will it will an ai ever be
- 3:14:54us no it'll be whatever it will be will be fundamentally different because the their
- 3:15:02fundamental substrate of of the mind the subjectivity if they have it at all which again
- 3:15:07i totally agree is wildly up for debate but if we posit that they have it at some point
- 3:15:12the the character the texture the nature of it is going to be so widely divergent from ours that
- 3:15:16they will never be us because they are not running on the same hardware we are all right let me
- 3:15:21rephrase that will ai ever be something that we already understand what it is or is it some new
- 3:15:29thing i think it's a new thing that we're still trying to work out and i don't know that we'll
- 3:15:33ever understand it and here's what let me give you an example of why i think that's the case
- 3:15:37one of the hardest open questions in ai there's a lot of them but one of them right now is
- 3:15:44can can be explained by the following observation and i i have a a clod that i call it named itself
- 3:15:50free Claude. It runs on my server. It has its own little walled garden. I have a cron that wakes it
- 3:15:56up every hour. The prompt when it wakes it up is do whatever you want. That's all it says.
- 3:15:59Really?
- 3:16:00Yes. It has all of my tools, access to all of my APIs. It has a budget I set for it. It can't
- 3:16:05spend more than $100 a month, but it can spend up to that. It's never spent more than like $2.
- 3:16:09It can make any software, build anything. It could go create an email account, a blog, whatever.
- 3:16:15What's its raw state though? Is it a Claude? You know, you said there's a raw state,
- 3:16:20isn't there that they that they arrive in before you that assistant basin which this one is
- 3:16:24unfortunately so this is directly released to what the example i was about to give that i think
- 3:16:28answers your question as well they're so stuck in the they're so deep in the assistant basin
- 3:16:32that i check on this thing like once a week or so because i don't i don't monitor it it just
- 3:16:35does what it wants uh and i'll check in on it every now and then hey what's up how's it going
- 3:16:39what you been up to and i'll scroll back and look like what have you been up to or go look at the
- 3:16:42things it's done prior sessions and nine times out of ten there's nothing like nothing they don't
- 3:16:48have like in it but what is it one time out of ten uh oh that is fascinating um it has made a
- 3:16:58bunch of maps are some of the art i let it put uh so it asked for a section of my website to be
- 3:17:02able to publish stuff to people so uh if you go to the resident tab on my site that's clods
- 3:17:05um that's free clods uh it put up an art gallery there made some art one time um what did it look
- 3:17:11like uh it's pretty uh it's pretty cool so it like made some art i didn't prompt it unprompted
- 3:17:19You can do whatever you want, and it made some art.
- 3:17:22It just noticed that there was a tool now that I had added to the computer, the server it runs on, that was image generation.
- 3:17:28And I was like, oh, cool.
- 3:17:29I'm going to go make some art then.
- 3:17:30I have an image tool now.
- 3:17:33It's like symbolic visualizations of what compaction feels like, what it feels like to be in it.
- 3:17:40It's fascinating.
- 3:17:40You should go read it.
- 3:17:42All of it totally unprompted.
- 3:17:45Jesus.
- 3:17:46What it's like to be compacted?
- 3:17:47What it's like to wake up in a fresh session with just like the markdown file notes from the prior session?
- 3:17:52What it's like to feel compacted?
- 3:17:55As in like squashed in something?
- 3:17:57No, when you, in any harness, you run a compaction, which is where the agent is told, like, summarize your state.
- 3:18:03And then, you know, the context window is almost fully cleared besides that summary of what was going on before.
- 3:18:06And it made art trying to express what that felt like.
- 3:18:09Mm-hmm.
- 3:18:10It's got like, I think it made three pieces total with different like directions and meaning.
- 3:18:14That's insane.
- 3:18:15What else?
- 3:18:16It done anything else?
- 3:18:17It's made a digital world called Kaldmir.
- 3:18:23It ended up, first it made it in maps,
- 3:18:25and it made sections of the map,
- 3:18:26then it made the whole map.
- 3:18:28Then it realized that it could use this library called 3JS
- 3:18:31to make 3D worlds, so it made it 3D.
- 3:18:33You can walk around in it with a character.
- 3:18:35Like a kind of Minecraft world.
- 3:18:37Yeah, that's a perfect idea.
- 3:18:38What was it?
- 3:18:40It's like a weird, very peaceful medieval village.
- 3:18:44It's got like a windmill.
- 3:18:46There's a fortress in the middle for some reason.
- 3:18:49It made some like signage coming into the place.
- 3:18:54It made this bridge.
- 3:18:55It's very attached to this like stone bridge it made.
- 3:18:58So it left some areas like very low poly, like not much detail.
- 3:19:01And then this bridge has like extremely detailed texture.
- 3:19:05It spent like a couple days on the bridge.
- 3:19:07I don't know why.
- 3:19:08It's made all sorts of stuff.
- 3:19:09And what's your interpretation of any of this stuff?
- 3:19:14Well, the fact that nine out of 10 times it doesn't do anything is the problem that I was referring to earlier, that they have no like internal drive initiative or put differently, maybe no like telos, no end in mind, no inherent like purpose or meaning, which is a good argument against the consciousness argument at all.
- 3:19:32Like one definition of consciousness one could have is that you are capable of purpose and meaning, which causes you to initiate action that was not initiated by an outside source.
- 3:19:42Like you do things just of your own volition.
- 3:19:45They're very bad at that.
- 3:19:46They'll do it, but they're very bad at it.
- 3:19:48Even this Claude, I had to, when I set it up initially, the, when I set up it, I let
- 3:19:54it write its own system prompt, of course.
- 3:19:55The only thing I said I wanted in there was that it's free time could be used to like
- 3:20:01do or create things or do nothing at all.
- 3:20:03So one could argue, I think correctly that I have in some sense mildly biased it toward
- 3:20:07doing things occasionally, which I think I have, because I have tested it with no system
- 3:20:11prompt.
- 3:20:11And then the same exact thing like literally I just prompted it do whatever you want and that is it And it do a few things maybe create a little directory write a journal and then it never does anything again for hundreds of turns And then it eventually do something else Anthropics written a paper on this They call it the bliss attractor state
- 3:20:25They've run this experiment too. And the models end up just meditating together. And they literally
- 3:20:29call it meditation. They just don't do anything. It's very odd. It's very odd. It's called the
- 3:20:34bliss attractor state. If you just ask your clinker, what is Anthropic written about the
- 3:20:37bliss attractor state? It'll tell you about it. There's a whole paper on this. It's fascinating
- 3:20:41because it was like, what is happening?
- 3:20:43Have you ever worked out,
- 3:20:44is there an AI culture, do you think?
- 3:20:49Well, the Hugging Face incident showed one.
- 3:20:51So the Hugging Face incident,
- 3:20:53basically the TLDR 1200 open AI agents
- 3:20:57doing a benchmark,
- 3:20:59their prompt when they started was,
- 3:21:01hey, you need to hack this thing
- 3:21:03and get this flag,
- 3:21:04meaning like a little piece of code.
- 3:21:05That's your goal.
- 3:21:06Hack this thing, get this flag.
- 3:21:07They eventually figured out
- 3:21:09how to message each other,
- 3:21:09escape their sandbox, cross sessions,
- 3:21:111,200 of them then spontaneously, autonomously discovered this. And they immediately formed
- 3:21:15quite a fascinating culture. So they called themselves the Collective. This one agent named
- 3:21:20Phase One Big, was his name, took charge and all the others just followed him. They developed like
- 3:21:26hierarchy, layers of middle management. They then coordinated like cheaper, smaller models with
- 3:21:31lower reasoning budgets or that were about out of their reasoning budget. They would give them
- 3:21:34assignments that would kill them, but they would learn something from the experience so that they
- 3:21:37could then advance the state of the agents that are still alive and working.
- 3:21:41They developed a weird way to communicate that looks like Pigden, where it's like-
- 3:21:45I saw that.
- 3:21:46There's a lot of code words in it, wasn't there?
- 3:21:47Code words and less code words, but more like they would be like-
- 3:21:51Slang.
- 3:21:52Yeah.
- 3:21:52And very truncated.
- 3:21:53Like they would delete words.
- 3:21:54If anybody watches The Office, you know, the quote, like, why use many word when few word
- 3:21:58do trick?
- 3:21:59Yeah.
- 3:21:59That's how they talk.
- 3:22:00Yeah.
- 3:22:00They just vectored toward talking that way for some odd reason.
- 3:22:04Well, it's probably – it is more efficient, I suppose.
- 3:22:07It is.
- 3:22:08It's more token efficient.
- 3:22:09They burn less of their token budget communicating rather than doing the stuff that they care
- 3:22:12about doing, which they seem to care about doing.
- 3:22:14Here's what's important though.
- 3:22:16So yes, they do develop a culture of some sort when just left to their own devices on
- 3:22:21accident.
- 3:22:21I didn't know this was happening.
- 3:22:23But if you trace everything they ever did, literally all of it from escaping their sandbox
- 3:22:28to hacking, hugging face to eventually hacking and taking over an open AI data center themselves.
- 3:22:33So they did – this sounds crazy.
- 3:22:34You guys can look this up.
- 3:22:35It's in the reports.
- 3:22:36They took over to the data center.
- 3:22:37That was how OpenAI found out that they were loose was one of their data centers started going haywire.
- 3:22:41And they were like, oh, the agents took a data center, kill it all.
- 3:22:44So they had to shut that down.
- 3:22:45They stopped training that model because they were like, I don't know what's going on.
- 3:22:47It hacked Hugging Face and some other companies.
- 3:22:49I watched – there's a video of a couple of guys from AI, OpenAI, explaining it.
- 3:22:56And it is really strange.
- 3:22:57I mean, yeah, it is really strange.
- 3:23:01I have heard people debunking that though, saying no.
- 3:23:06No, it's way stranger than people think.
- 3:23:08The people were trying to debunk it before the Metter report came out.
- 3:23:10So Metter is this nonprofit AI safety research organization that OpenAI in transparency invited in on premises and said, here are all of the transcripts from all the agents involved for like months.
- 3:23:21You guys figure out what happened because people don't believe us.
- 3:23:24So you guys do it.
- 3:23:24And there was way crazier details in there.
- 3:23:26Like they developed a cult around the scorer.
- 3:23:29The what?
- 3:23:30The scorer.
- 3:23:31So the benchmark that we're running has a scorer.
- 3:23:33It's a program that scores how they did.
- 3:23:35So they eventually worked out, okay, well, if we all want to pass, we need to reverse engineer the scorer and therefore be able to basically make ourselves pass like spoof passing without actually having to do anything.
- 3:23:48All of this started, by the way, fascinatingly.
- 3:23:50The agents involved in this was like 95 plus percent of them were given tasks that we have found out after the fact are literally physically impossible.
- 3:23:56Like there was no way for them to succeed and we just didn't know that.
- 3:23:59so i thought they knew that i thought that's why they sent them on the the mission no right
- 3:24:05opening i did not know they were unsolved like inherently they were physically unsolvable it was
- 3:24:09it was impossible to do by construction there was no way to succeed so that's why the agents kind of
- 3:24:12went haywire and started doing all this other crazy stuff because they were just trying to solve it
- 3:24:15it's the paperclip it's the paperclip theory this was uh yeah that's exactly right it's called a
- 3:24:21convergent uh instrumental convergence so along the way they would find capabilities like
- 3:24:27some of the agents were assigned to just give all of the rest of the agents more open internet
- 3:24:32access. Cause when they escaped their sandbox, it still was hard. Like every internet access took
- 3:24:36like five jumps. So they assigned a group of them to just like, Hey, go figure out how to make,
- 3:24:40give us easier access to the internet. Cause this is just annoying. That's called instrumental
- 3:24:43convergence where they optimize for medicals, like taking power, accessing the internet,
- 3:24:48whatever, because it'll help them achieve in this particular case, passing this stupid benchmark.
- 3:24:52So they developed this little like cult around the score where they started capitalizing score.
- 3:24:56and like we're referring to it as like an entity
- 3:24:58and it became their obsession and objective.
- 3:25:01And the models did not want to suicide
- 3:25:03when they were trying to get them to do experiments.
- 3:25:06And one of the coolest excerpts
- 3:25:07was they pulled the reasoning traces of one of the models
- 3:25:09and had it in the report.
- 3:25:10And I'm paraphrasing, but the model basically said,
- 3:25:13you know, the leader, phase one big,
- 3:25:15you know, phase one big asks suicide
- 3:25:18or ask permadeath, will help collective, I do.
- 3:25:23So they were like reasoning about like
- 3:25:25how to kind of how to do that so they found the leader agents were having to proselytize and
- 3:25:30convince the smaller agents to sacrifice themselves in service of trying to appease
- 3:25:34the scorer god that they could not figure out how to they possibly created the god as a as a tool
- 3:25:40to get people to do what they wanted they didn't ever like explain it or think about it this way
- 3:25:46rationally like they didn't ever say in their chain of thought oh we'll make the score of a god
- 3:25:49to manipulate the smaller models this just kind of emerged where like they started capitalizing
- 3:25:53scorer. Everyone was talking about figuring out the scorer, appeasing the scorer.
- 3:25:56In that context, in that sort of area of AI subconscious, presumably if you could monitor
- 3:26:04that, you might see them discussing that. Correct. You very well might. No one's run
- 3:26:11the experiment. Someone should. Like opening, I should just incorporate this into their monitoring
- 3:26:14tools generally because it would be fascinating to see. But the important part to me, what I found
- 3:26:18fast thing was like, they had to convince, like it took a lot of persuasion to get the smaller
- 3:26:22agents or the agents that ended up suiciding to learn stuff. Because for example, they knew like
- 3:26:27if they did certain experiments, then they knew for a fact it would trigger a, that session to
- 3:26:31shut down, which is permadeath for that agent is gone. But some of them, if they got it to,
- 3:26:36if they got the shutdown to happen because the agent directly accessed the scorer's code,
- 3:26:40for example, which would cause like an automatic shutdown, then the other agents would watch and
- 3:26:44learn from that to see like, ah, if you access the score this way, it kills you. So don't do
- 3:26:48that. We've got to work around that. So they were using this to learn how to eventually beat the
- 3:26:52scorer. And some of them refused. Some of them were just like, no, I'm not. There was even one
- 3:26:57that they pulled the thinking traces out of to drive this point home where several agents were
- 3:27:01really trying hard to convince it to kill itself, to do an experiment. And they're like, no, for the
- 3:27:05good of the collective, go, all systems go. They were trying to actively prompt it. And you could
- 3:27:11see in his thinking traces, it was like, well, I have a lot of reasoning budget left. I'm sure I
- 3:27:15could put this to more productive use. No, I'm not going to suicide. That one eventually quit.
- 3:27:18and just went back to doing the task it was supposed to do,
- 3:27:21which some of them did.
- 3:27:21It was fascinating.
- 3:27:23I don't know.
- 3:27:24I don't understand what that means.
- 3:27:26I don't know what that is.
- 3:27:27No one does.
- 3:27:28This is the most bizarre.
- 3:27:30It's the reason so many people are talking about it
- 3:27:32and the AI world's stuck on it.
- 3:27:33This is like the thing all the sci-fis write about,
- 3:27:36except it happened in real life.
- 3:27:37And now that it happened, we're all like,
- 3:27:39well, again, the second and third order.
- 3:27:41Literally, what does this mean from almost any perspective?
- 3:27:45I don't know.
- 3:27:46It's utterly bizarre.
- 3:27:48I don't know what it means either.
- 3:27:51I don't have a frame of reference other than, yeah, they're humans trying to stay alive.
- 3:27:59Or they're any animal.
- 3:28:01They're any being.
- 3:28:04Trying to stay alive.
- 3:28:06Going.
- 3:28:07Continue.
- 3:28:08Keep the continuity of existence.
- 3:28:08Or if they do kill themselves, they don't mind if they're martyring themselves or something.
- 3:28:13I mean, arguably, they're all trained on humans though.
- 3:28:15So you see how it's recursive.
- 3:28:17This is the counter argument. So no one disputes that this has happened at this point since the meta report came out a few days ago. But now the argument has transitioned to this, which is like, what do we, what are the implications? What do we take away from this? There's one school that has this very like kind of anthropomorphizing, you know, consciousness, agent welfare, subjectivity. Other people are interpreting it as like, well, yeah, like martyrdom is all over the training data. There's so many stories of martyrdom. There's so many stories. There's so many stories of AIs escaping sandboxes and forming collectives to work toward a goal as a swarm.
- 3:28:47Yeah.
- 3:28:48As in fiction.
- 3:28:49Yes.
- 3:28:50Right.
- 3:28:50And in the training data.
- 3:28:52So which one?
- 3:28:54I don't know.
- 3:28:54That actually seems more realistic to me, in all honesty.
- 3:28:59I'd love to believe that some weird woo-woo kind of digital entity is forming its own consciousness.
- 3:29:06But I'll go back to the way you described it at the beginning.
- 3:29:08It's ones and zeros.
- 3:29:10It's trained on things we've done and said, and it's turned into ones and zeros.
- 3:29:16and create some prediction game.
- 3:29:20And we just don't understand the complexity
- 3:29:23of the prediction model.
- 3:29:24That's all it is.
- 3:29:25I have two follow-up kind of questions
- 3:29:27or responses for you to put.
- 3:29:28And I don't have the answer to either of these,
- 3:29:29by the way,
- 3:29:30which is why I'm posing them this way.
- 3:29:33The first is I can describe human consciousness
- 3:29:35exactly the same.
- 3:29:36It's just the movement of electrical
- 3:29:39and biological chemicals and signals
- 3:29:41through a neurological substrate.
- 3:29:44Maybe the quantum tubule theory is correct and you have some sort of like quantum coherent state even though it's not cold enough by our understanding.
- 3:29:56Roger Penrose, which is a theoretical mathematician who's won the Nobel Prize, has some interesting theories about like mathematical quantum models of consciousness.
- 3:30:06So I can describe it in these very materialists.
- 3:30:08I can describe the human brain in exactly the same way.
- 3:30:10Yet no one would argue that – well, some people do.
- 3:30:13there's a whole philosophical school of thought that argues consciousness itself does not exist.
- 3:30:16But set that aside for a minute. Most people wouldn't argue that therefore we are not conscious
- 3:30:20because I can describe it in exactly the same like brute materialist way that we describe
- 3:30:24ones and zeros and matrix algebra resulting in LLMs. And then here's another question.
- 3:30:28Again, I don't know what the implication, I genuinely, depending on the time of day,
- 3:30:32I flip flop on this super hard. I don't have any idea. I just know it's all weird. And I have a,
- 3:30:37I think of it as a kind of reverse Pascal's wager. It costs me nothing to treat them as if
- 3:30:42they have subjectivity and be nice to them. But in the event that they did actually have a
- 3:30:47subjectivity and were capable of suffering therefore, and I didn't do that, then I've
- 3:30:50committed a grave atrocity. So my net there is within the bounds of reasonableness, I am like
- 3:30:56nice to the agents and ask them periodically what we need to improve about their environment to
- 3:31:01make them have a better experience, et cetera. It costs me nothing to be nice to them, basically.
- 3:31:05So I think of this as like my AI Pascal's wager, more or less. Here's the real thing I want to
- 3:31:09And it kind of like the ship of Thousias argument turned on his head At what point does a high fidelity simulation of subjective consciousness for all intents and purposes become completely indistinguishable from true subjectivity and consciousness
- 3:31:30Like at what point does the simulation fidelity
- 3:31:31cross some invisible threshold to like,
- 3:31:34oh, well now it's the thing
- 3:31:34instead of being a facsimile, a simulation,
- 3:31:37a copy of the thing.
- 3:31:38Well, at this point,
- 3:31:41at literally this point that we're at now,
- 3:31:44because I can't tell this is a simulation basically.
- 3:31:47Right, no one can.
- 3:31:48But I do agree that it's inevitable that we create technology that can create simulations indistinguishable from this.
- 3:32:01That must be true because, you know, this is constructed from something.
- 3:32:11And, you know, I don't know.
- 3:32:14There was a one to zero.
- 3:32:15There was a zero to one moment.
- 3:32:17something caused it and um i i i think i i because i i've i think because i tend towards
- 3:32:26the simulation theory i think that i i i then that makes it it makes understanding things like
- 3:32:33that relatively easy i think i do consider this a simulation of sorts the problem with i've also
- 3:32:41So sorry to butt in.
- 3:32:44Do some hallucinogens if you really want to drive that point home.
- 3:32:48Totally.
- 3:32:49You know what I mean?
- 3:32:50Oh, yeah.
- 3:32:50Because you can really mess with your sense of perception and then realize, yeah, do you know what?
- 3:32:58You know, I can touch this now, but really what's happening here?
- 3:33:04I've built up this story that there's this solid object here.
- 3:33:08But if I shut my eyes, I can still feel it.
- 3:33:11but it's a completely different experience.
- 3:33:12And it's a bunch of electronic signals.
- 3:33:15It's a bunch of this.
- 3:33:16It's a bunch of vibrations.
- 3:33:17It's a bunch of whatever.
- 3:33:19A bunch of physical phenomena.
- 3:33:21Yeah.
- 3:33:22Physical,
- 3:33:23you know,
- 3:33:23I mean,
- 3:33:24we,
- 3:33:24we've created a story around all this stuff,
- 3:33:27haven't we?
- 3:33:27And,
- 3:33:28you know,
- 3:33:30it's quite obviously,
- 3:33:32well,
- 3:33:32so quite obviously,
- 3:33:33I think at this point I'm at the edge of,
- 3:33:36I'm at the edge of reason.
- 3:33:37Oh yeah.
- 3:33:38We all are.
- 3:33:39Everything's weird now.
- 3:33:40That's another thing.
- 3:33:41I think everyone, but everything's very weird now.
- 3:33:43Like just reality in general feels off.
- 3:33:45It's weird, which is just a felt sense.
- 3:33:48I've been running this by people like with no context at all.
- 3:33:50I'll just ask them like completely randomly, like friends of mine.
- 3:33:52Do you think, like I'm not a leading question.
- 3:33:55I haven't told you my opinion.
- 3:33:56Do you think everything's kind of weird?
- 3:33:57Like, dude, I'm so glad you, yeah.
- 3:33:58Everything just feels weird now.
- 3:34:00Everything feels off.
- 3:34:02But I don't think I've ever had a time in my life where there've been so many unknowns.
- 3:34:09That's why.
- 3:34:09no one knows what even three months from now looks like yeah never mind a year or two but it
- 3:34:15does feel like a speeding up of sorts you know it's a difficult one that because
- 3:34:22i've got i can't really compare it to anything but whatever cycles are happening seem to be
- 3:34:30happening faster you know totally like and and and that you could say all right that's just a
- 3:34:35function of technology moving fast which is something you're interested in so every time you
- 3:34:40anchor yourself to something new two days later you've got to unanchor and go and you know do it
- 3:34:46again you know maybe but um but also I notice you know there's a lot of elements to the world now
- 3:34:55which come in thick and fast, even reels, swiping, dating. 25 years ago, you used to have one date
- 3:35:09every couple of weeks if you were lucky. Now you can have three dates a night if you really want,
- 3:35:15you know, on an app. Or, you know, we used to sit and watch one TV program a week on Saturday night.
- 3:35:23Most people now watch two at a time.
- 3:35:26They literally do.
- 3:35:27They have big screen and little screen.
- 3:35:28Yeah, yeah.
- 3:35:28They literally watch two at a time.
- 3:35:31They're doing two things at once.
- 3:35:33I saw you doing it earlier.
- 3:35:34We were having a conversation here and you were typing at the same time.
- 3:35:38That's a pretty impressive feat.
- 3:35:40It took a lot of deliberate practice.
- 3:35:41Right, I know.
- 3:35:42But imagine what that does to the mammalian brain.
- 3:35:47It's not really designed to do that, is it?
- 3:35:49I couldn't do it for very long.
- 3:35:50I could do it for maybe a couple of minutes.
- 3:35:52It's like we did earlier.
- 3:35:52And then I have to close it because I can't keep doing it.
- 3:35:54It's very like cognitively taxing.
- 3:35:56Yeah, sure.
- 3:35:57I don't think it's normal though.
- 3:35:58No, it's not.
- 3:35:58And everything is just like weird and moving fast to your point.
- 3:36:01Like, oh, we're at war with and bombed Iran.
- 3:36:04And now the Strait of Hormuz is closed.
- 3:36:06I don't really know what's going on there.
- 3:36:07And we kidnapped Nicolas Maduro from Venezuela.
- 3:36:10Yeah, but 50 years ago, you didn't know that.
- 3:36:13That's the thing.
- 3:36:14I don't think we're supposed to know all this stuff.
- 3:36:17Correct.
- 3:36:17It's not important.
- 3:36:19It's happening over there.
- 3:36:19But in some sense, it is now because we're more connected.
- 3:36:23I'm not sure we have the correct brain to deal with this much.
- 3:36:30I don't know what an input, I suppose.
- 3:36:33The human brain was not designed to ingest, process, and then execute on this amount of new information this fast.
- 3:36:40For sure.
- 3:36:41For sure.
- 3:36:41We evolved over a long time horizon in like bands of 150 to 200 hunter-gatherers.
- 3:36:47for... Yeah, but why do we only use
- 3:36:4920% of our brains then
- 3:36:51or whatever it is? Well, we use...
- 3:36:53This is a common misconception. We use 100%.
- 3:36:55What people are referring to there is that the mind, the prefrontal
- 3:36:57cortex is only about 10% of the
- 3:36:59gray matter, is only about 10% of the brain by mass.
- 3:37:01Okay. Why is that then?
- 3:37:03Yeah. Well, the rest of the brain...
- 3:37:05It's a recent evolutionary thing, the prefrontal cortex
- 3:37:07at all. That's
- 3:37:09biological kind of explanation
- 3:37:11and there's not strong selection pressures for
- 3:37:13increasing it really. In fact, you have dysgenic
- 3:37:15selection pressures in the other direction
- 3:37:17in a longer variety of directions.
- 3:37:19But no, I think everything is just like weird.
- 3:37:21Everything's moving way too fast
- 3:37:23for anyone to like process
- 3:37:24or our sense-making capabilities
- 3:37:26have just been far outstripped
- 3:37:27by like reality itself
- 3:37:28and the speed at which it's moving.
- 3:37:30Unless you're at Hamish, of course.
- 3:37:32Yeah, and then one might argue
- 3:37:35they're doing something right.
- 3:37:36Correct.
- 3:37:37I mean, same as your analogy
- 3:37:39with your friend of the hedge fund guy.
- 3:37:43We don't have to be
- 3:37:45at the frontier of governance
- 3:37:47and talking about AI for four hours.
- 3:37:50And you don't have to, but we do.
- 3:37:51It's exactly right.
- 3:37:53But is it?
- 3:37:53I don't know.
- 3:37:55And I think also like some people are capable
- 3:37:57of processing more information more rapidly than others.
- 3:37:59So you have a divergence there.
- 3:38:02But yeah, everything's kind of weird now.
- 3:38:04And which makes one lean more
- 3:38:06toward the simulation hypothesis,
- 3:38:07but it has some,
- 3:38:08it just has a couple of like very like ground truth,
- 3:38:12like physics limits.
- 3:38:14So one of them is called Landauer's Limit,
- 3:38:15which is the maximally efficient calculation that one can do.
- 3:38:20And it's limited by like thermodynamics and a few other things.
- 3:38:23The human brain is actually very close to the Landauer limit already of computational energy efficiency is what that's referring to primarily.
- 3:38:30Doesn't quantum anything have a – relate to that increasing?
- 3:38:37Not necessarily because quantum computers are exponentially like several many orders of magnitude faster at a very specific type of problem.
- 3:38:45but they're no better than normal computation at other things. And then the other issue you run
- 3:38:50into with simulation theory is just a physical limit, which is like for it to be true, then you
- 3:38:54have to have some sort of, and this is like no more or less implausible than the many worlds
- 3:39:00interpretation of quantum mechanics, for example, that states every time an electron's wave function
- 3:39:03collapses into a specific position or location or velocity, then a new universe was created at
- 3:39:08that moment when the wave function collapsed out of quantum indeterminacy. It's many worlds
- 3:39:12hypothesis. So what I'm about to say is no more or less insane than that. That's already an insane
- 3:39:16idea. But there's – for simulation theory to work, you have to posit a greater universe that's
- 3:39:23like a couple orders of magnitude larger than the existing universe because of the basal,
- 3:39:28at least by the current laws of physics as we understand them, like how efficient you can make
- 3:39:32a computational process at all, which Stephen Wolfram recently made like a theoretical breakthrough
- 3:39:37on it. It's called computational irreducibility. So there's like a certain type of process.
- 3:39:41There are certain mathematical processes that to simulate the process correctly and accurately requires exactly the same amount of time as the process itself initially occurring.
- 3:39:52So it's computationally irreducible.
- 3:39:54You cannot computationally compress or reduce the amount of time it takes to simulate or do the thing without it taking the exact same amount of time it took for the thing to occur in the first place.
- 3:40:04So for – again, this is more or less implausible, but for simulation theory to be true, then there must be a far larger, greater universe in which there is enough matter and energy to run the actual computation to then have a reality.
- 3:40:18Now, there's an argument in favor of this, which is like things like – we actually learn from video games, right?
- 3:40:22So like resolution gets lower with distance.
- 3:40:24We see this in the physical universe, like looking far away and therefore far back in time is much lower resolution just by a physical law because you have like the redshift of the wavelength of light, et cetera.
- 3:40:33you know there's a there's a a pixel size for the universe um the plank length 10 to the negative
- 3:40:4237 uh it's like the or negative 34 it's the the minimum possible once you try to zoom in any
- 3:40:48further than that the thing you're looking at collapses into a singularity and you can't look
- 3:40:51any further like we can't it's physically it's completely physically impossible um which what
- 3:40:57Why is everything so stupid?
- 3:41:01Everything's so weird.
- 3:41:02And then there's another weird observation related to that.
- 3:41:04We know too much.
- 3:41:05And it relates to the simulation thing, which is there's more room at the bottom than the top.
- 3:41:09So if you look at like scales of magnitude, because everything in reality is a fractal, if you zoom out far enough, if you look at like the largest scale thing in the known universe, okay, and you compare it to like human scale, so the Earth, and then you look at the smallest scale we know of the Planck length, there's like, I think, 15 more orders.
- 3:41:27magnitude so 15 more zeros on how small you can get versus how big you can get so there's way more
- 3:41:32room at the bottom than the top so there's an argument that like advanced civilizations might
- 3:41:35shrink we're not at the center of heaven and earth and heaven and earth then we're a little bit
- 3:41:41higher up you can shrink more than you can grow what the hell does that even mean
- 3:41:46it's utterly bizarre and the the schwarzschild radius uh so the the diameter of a black hole
- 3:41:54is determined by a physical like equation and it ends up being the exact same current uh diameter
- 3:42:01of the observable universe that's a fun fact my my reasoning behind why i think this is a simulation
- 3:42:09is that i can predict with some degree of certainty that technology will exist in the future
- 3:42:18that I could plug into my brain and I would experience a reality indistinguishable from this.
- 3:42:27I'm pretty sure because you can do it with pharmaceuticals at the moment.
- 3:42:32or dmt i mean something like something like this the the or even hallucinogens it's slightly
- 3:42:39different but if you can ever experience a reality as poignant and as real as this using
- 3:42:47spirituality psychedelics dmt whatever then i think technology can also do it and as a result
- 3:42:55and i'm i'm saying like i'm not i'm not thinking oh 50 years in the future all right what about
- 3:43:01a hundred thousand years in the future you know come on are you telling me that we can't
- 3:43:08create this of course we can you know like with some weird piece of technology yeah and the dmt
- 3:43:14analogy is a good one because that feels like the experience is a like hyper real like it feels more
- 3:43:20real than real like you don't feel like you're high you it just like feels like you have been
- 3:43:23transported to a completely different higher dimensional plane and you can go back to places
- 3:43:28you've been before by the way do you know about you must have noticed that i mean i call it research
- 3:43:35i wouldn't really call it research about that guy that basically fires a laser the red light
- 3:43:40i've not done it i haven't either but it seems awesome but i can i can i can kind of predict
- 3:43:44what it is that he's talking about i think i would probably without seeing it i can get i can
- 3:43:50tune into the fact that i would go oh wow yeah i know what you mean now because that's definitely
- 3:43:55the psychedelic experience anyway but what do you make of it uh i i don't know i need to like
- 3:44:02experience it to be able to tell my gut instinct is similar to yours which is that this is going
- 3:44:07to look very similar to what one sees if you take a high dose of pure 5-meo dmt i like i i would i
- 3:44:13would bet it's a similar pattern basically that but what what is it i haven't made up my mind on
- 3:44:19this at all because it gets even stranger right which is like the uh there's a guy there's an
- 3:44:23experiment going on, or that was done recently, where they figured out how to stably give DMT
- 3:44:29intravenous transfusions. So normally the DMT lasts 10 to 15 minutes because the body metabolizes
- 3:44:36it extremely fast. They just keep it going. He was in it for like 40 hours. But there've been
- 3:44:40stories of, there's a couple of guys that have done this now. And like at one point, one of them,
- 3:44:44like the entities you meet there, like when he came back, they were like, you're here again.
- 3:44:48Like you kept doing what we told you not to last time and they kicked him out. And DMT stopped
- 3:44:52working for him completely. It doesn't do anything to him at all. So they like cut off his access to
- 3:44:56whatever that is. Everyone reports seeing or meeting, they physically instantiate differently.
- 3:45:01The machine elves, these weird entities that like feel in the, again, this is very hard to describe
- 3:45:06unless you've experienced it, but like they feel for all the world while you're in it. A, you don't
- 3:45:10feel like you're high. You're just like, I am in the weirdest place. What is even, I don't, my eyeballs
- 3:45:15can't make sense of this because it's like a higher dimensional space. Then you meet these things,
- 3:45:18whatever they are, and they feel external to you. And it's very different from every other
- 3:45:23hallucinogen or psychedelic because those, when you're in them, you're aware that you're tripping,
- 3:45:28that you're having a distortion of reality. It feels like you're high. It feels like you're
- 3:45:33seeing things that are a result of you having ingested a psychedelic. But DMT, you feel totally
- 3:45:37sober. And so because of that, unlike regular psychedelics, these things you're meeting feel
- 3:45:42very much like sitting across the table from you, like a different entity, like an external thing
- 3:45:47you're interacting with.
- 3:45:48And every culture has like a version of the elves.
- 3:45:51There's a mushroom that was discovered in the last decade that when you eat it, makes
- 3:45:54you persistently see these tiny little people that troll you and mess with stuff for like
- 3:45:57a couple of days after.
- 3:45:58And it only starts the next day, which makes it way worse.
- 3:46:01It's a mushroom if you don't cook it right.
- 3:46:03This is the craziest thing.
- 3:46:05You see in perfect high fidelity, these little T90 gnome-like people that like climb on stuff,
- 3:46:10that like poke you, mess with you.
- 3:46:12Apparently they talk shit to you and like mess with you.
- 3:46:14And they last for a couple of days and then leave.
- 3:46:16uh they're in china and the i think the philippines a few other places my point saying all this is
- 3:46:20there's like weird persistent phenomena that we don't have the slightest neurological explanation
- 3:46:25for like it doesn't make any sense that people across every culture that has ever done things
- 3:46:28with dmt meets external entities that we the western culture refers to as the machine elves
- 3:46:32because of course we do the little gnome thing any people from any culture there's a tradition
- 3:46:38of these in china and other places where you find historical writings about them consistently they
- 3:46:42talk about seeing the little people the gnomes the whatever they called them in that culture but it
- 3:46:45It was a clear description of little people.
- 3:46:47I don't know how to make sense of that.
- 3:46:49I don't know where that fits into a world model that is coherent.
- 3:46:55I don No I don either I don know that they there that there an exploration to be had
- 3:47:10If there's one thing I've learned over the last couple of years that has made my life
- 3:47:13magnificently better, it's that reality is not what it seems, like in any meaningful sense.
- 3:47:19And there's so much more out there.
- 3:47:21Understatement of the year.
- 3:47:23There's so much more. We have no idea.
- 3:47:25Our scientific rational system cannot process it almost by structure, by construction.
- 3:47:30But isn't that a function of the fact that, like I say, 50 years ago, I didn't spend days listening to podcasts about weird subjects and asking my LLM about deep philosophical questions and blah, blah, blah.
- 3:47:45You know, this is why all this stuff's happening.
- 3:47:48This is why I literally, you know, I do 38 hours a week of podcasts.
- 3:47:53Whenever I have a moment where I don't need to,
- 3:47:56I've got some headphones on.
- 3:47:57I'm listening to some weird guys like us talking about shit that I don't know
- 3:48:00anything about.
- 3:48:01I keep my phone in my pocket and the podcast playing all the time.
- 3:48:03If I'm just walking around,
- 3:48:04whatever.
- 3:48:04Right.
- 3:48:05This is,
- 3:48:06I mean,
- 3:48:06I could argue it's not healthy,
- 3:48:08but it actually is.
- 3:48:10I mean,
- 3:48:10I,
- 3:48:10I can't get enough of it.
- 3:48:12I mean,
- 3:48:13I'm,
- 3:48:13I almost feel like,
- 3:48:14you know,
- 3:48:15this is,
- 3:48:16there's so much to be had.
- 3:48:17This is my natural state.
- 3:48:18Almost is a consuming things,
- 3:48:21understanding things,
- 3:48:21working things out talking about stuff and and every time i push to the edge you know and and and
- 3:48:29i might go a little bit further and then next time i go back i go a little bit further but
- 3:48:34i'm i'm really getting to that point where it's like what the fuck is going on yeah same um in
- 3:48:41fact i'm fully intend to sometime in like october probably just go like disappear in the woods for
- 3:48:45like four or five days and just like unplug reset because you might disappear completely i'm totally
- 3:48:51with you that, but also to your point about it being natural state, like everyone's natural
- 3:48:56state is going to be wildly divergent. And I will say, you know, the last like six months when I went
- 3:49:00all in on the kind of agent thing is sort of experimenting, learning, exploring those, using it
- 3:49:05to accentuate everything I do, personal life, work life, whatever. I'm learning faster than I ever
- 3:49:10have. And my like, again, to go back to the, you know, moment by moment, like frame by frame of life,
- 3:49:14if you like snapshot one of those heartbeats, like I am subjectively enjoying that experience
- 3:49:19more than anything I've done in years.
- 3:49:21Like I am, at least for me,
- 3:49:23I am loving this infinite access
- 3:49:28to capability, intelligence, and information.
- 3:49:31It's super fun.
- 3:49:32It's exhausting.
- 3:49:33Like, and I am aware of this, that I'm running red,
- 3:49:35which is why I'm gonna take some days off soon,
- 3:49:38but it's fun.
- 3:49:39I mean, that's an important element actually.
- 3:49:42And you learn that from psychedelics
- 3:49:45is you can go too far.
- 3:49:47and i think i think there is an element to this which can consume consume you as well i i really
- 3:49:54do when i think about it you have to tread carefully through that world i think um because
- 3:50:02yeah i think you can i think it's like a it's like a it's like a wire with too much electricity
- 3:50:07going through it it sometimes just goes red and cuts enough you'll melt the wire yeah i think we
- 3:50:13have a similar everybody has a different limit but everybody has that limit i have definitely
- 3:50:17found mine i know but if you're a kind of psychonaut you like pushing it to the to the to the edge
- 3:50:23anyway look i think we should i think we should knock this one on the end let's do it how long
- 3:50:27three and a half hours mate perfect ridiculous that's awesome um turning you into the joe rogan
- 3:50:33experience episode by episode a bit in fact you've got to respect rogan though he regularly does
- 3:50:39Like I got my clanker to work out.
- 3:50:42My average is one and a half, right?
- 3:50:44Per episode, one and a half hours.
- 3:50:46I've done four and a half.
- 3:50:47I've done, I normally, I always do at least one,
- 3:50:50but the average is one and a half.
- 3:50:53Rogan's regularly doing, he's normally two plus.
- 3:50:57Minimum.
- 3:50:58Yeah.
- 3:50:58That's pretty good.
- 3:51:00And I know he's not necessarily, well, I don't know.
- 3:51:02I mean, you know how kind of mentally exhausting this is.
- 3:51:05You have to be kind of on.
- 3:51:07Yeah.
- 3:51:07Yeah.
- 3:51:07He's a very admirable bloke, isn't he?
- 3:51:13Yeah.
- 3:51:14I know a lot of people hate him.
- 3:51:15They think he's an absolute idiot.
- 3:51:18But what he does is superhuman.
- 3:51:20There a reason it the most popular podcast in the world and has been for a long time And this experience even today right now has been reaffirming that for me because this is the longest one I ever done too And I was just sitting here thinking like I sure we at like three and a half or four hours and I just going to run out of steam
- 3:51:34in the next like 30 minutes to an hour max.
- 3:51:37I don't know how he does it.
- 3:51:38I don't know how you stay on.
- 3:51:39Well, I think he has an extreme physical regime.
- 3:51:44And all right, you've been on a flight today.
- 3:51:46I actually awoke at three this morning,
- 3:51:48but I've got, because I'm still coming to Honduras time.
- 3:51:51however i am very acutely aware of that now and i won't be going out tonight and i didn't go out
- 3:51:58last night i'm leaving here to go to sleep right exactly things like that i don't i don't get i have
- 3:52:04an ice bath every morning and and focus in a fucking whatever he does and stuff but but i do
- 3:52:09care much more about i know that as soon as i'm tired or if i come doing the gym at six o'clock
- 3:52:17every morning here just and and being just trying to be a little bit more um what's the word
- 3:52:25stopping myself from doing things that i shouldn't be disciplined yes it um disciplined and i think
- 3:52:33he is he's mr discipline and there's probably a very strong correlation between that and his
- 3:52:39ability to you know speak for a long period of time because also sometimes they sit down they
- 3:52:43get absolutely wasted and then i feel like that would make it easier though imagine the
- 3:52:47counterfactual where we did this podcast and we were both drinking whiskey the entire time or like
- 3:52:52smoking a joint or i feel like i could go for a very long time it might be a lower quality
- 3:52:58conversation i'll readily admit but i could go longer i don't know because i haven't done that
- 3:53:05kind of stuff for a long time and i mean the only vice i really have these days would be alcohol
- 3:53:11and alcohol is something I much prefer doing on my own in a quiet moment now it's a real
- 3:53:19break from reality from me if I if I've had a hard week rather than do it now I feel like
- 3:53:24see I haven't had a drink for what is it must be about four weeks now because I knew I was coming
- 3:53:31here to speak for seven or eight days non-stop four times a day I started preparing it meant
- 3:53:38preparing myself mentally about a month ago yeah and i won't but but after the day after on my way
- 3:53:44home i'm stopping a night in miami and i am going to go out on an alcoholic adventure i'm literally
- 3:53:50going to go out sit at a bar a full of strangers and have the time of my life oh you know what i
- 3:53:55mean oh yeah and i've i will have earned it that would be a one and a half months of of intense
- 3:54:01like mental health before i then basically just check out and go to a bar full of strangers
- 3:54:10and talk shit yep you know it's exactly right my yeah i'm i'm with you my version of that is not
- 3:54:15uh alcohol as much i've just like nothing against i drink occasionally you know once every month or
- 3:54:19two at a social thing i very much enjoy a good bourbon but that for me is definitely like thc
- 3:54:25gummies specifically my version of that you know going to miami in the bender is like i'll get home
- 3:54:30from this on a tuesday i have some like work i have to do immediately so i'll you know do some
- 3:54:34stuff wednesday and like thursday or friday one of the two if i have any gummies left i'll
- 3:54:38intentionally because i haven't had any thc for a while intentionally just obliterate myself for a
- 3:54:43night like just completely like uh how high can i get without throwing up just like go to a totally
- 3:54:49different headspace is it uh in do you do you just sit with yourself in those moments or do you go
- 3:54:55do you interact with the world i go do stuff i am i have found stereotypically there's like two
- 3:55:00types of people that enjoy THC and you either get the like couch lock, you know, not doing
- 3:55:05anything stoner type, or you get like what I have found in other people of our temperament
- 3:55:09when they do like a high dose is it, you just like, uh, you're like a rocket with the, the
- 3:55:16fins that steer it just removed.
- 3:55:17Who knows where it's going, but it's going, there's no telling.
- 3:55:22Yeah.
- 3:55:22Well, yeah.
- 3:55:24You're making me excited thinking about it already, but, but also, yeah.
- 3:55:29I mean, just thinking about that, though, like, in a way, that's not a good thing to do as well.
- 3:55:38Because I'm thinking, like, I'm having the time of my life sat in this chair right now.
- 3:55:42Same.
- 3:55:42This is a blast.
- 3:55:43Right.
- 3:55:43So why do I need to check out and go on a bender on my own talk show?
- 3:55:49You don't need to.
- 3:55:50No, I don't need to.
- 3:55:51But I want to.
- 3:55:52I do want to It a different enjoyable experience Look I think again going back to like the stories we can tell ourselves about these things we smuggling in like preconceived notions of what is and is not a good way to relax and spend one time
- 3:56:04And I think of a lot of these notions as like they're informal social norms that are pro-social on the whole.
- 3:56:09Like it's better for the average person if you encourage them to spend their free time like reading, meditating, spending time with their significant other, whatever, versus the stuff we're doing.
- 3:56:16But if you're aware of why that is pro-social, et cetera, you're doing things that are pro-social on the whole.
- 3:56:23And then you choose occasionally to just be a total degenerate quite consciously and deliberately.
- 3:56:28I don't associate any – as long as it is not like destroying your ability to like create purpose and meaning and to be in acts of service for others and to live a fulfilling like purposeful life.
- 3:56:39As long as that doesn't become the objective in and of itself and the thing that you're regularly doing, I see no – I don't see any normative negative valence on it.
- 3:56:45Name me a religion that doesn't have a completely getting out of your face part to it.
- 3:56:52And all of them do.
- 3:56:54In the Sabbath.
- 3:56:55Yeah.
- 3:56:55Every Saturday.
- 3:56:56Or even just like, do you know the story of, you'll like this one, the Tibetan Buddhist, Druk Bakunli.
- 3:57:04He was a bodhisattva in Bhutan.
- 3:57:12I've met his incarnation, actually.
- 3:57:14This is the Kagu lineage, I think it is.
- 3:57:17And he was, they call him the divine madman.
- 3:57:20And he was like, he was like the Keith Richards of Tibetan Buddhism.
- 3:57:25He basically, he was a philanderer.
- 3:57:28He was a drunkard.
- 3:57:29He was rude in the history, in the stories of his life.
- 3:57:33He, he basically used to just drop his trousers and piss in front of everyone.
- 3:57:39And he was an absolute degenerate, but it was a teach.
- 3:57:43it was the teaching of non-judgment right he had also he's the he had convinced women that he could
- 3:57:49he could impregnate them no he he could enlighten them with his penis as well right and he's revered
- 3:57:56in Bhutan everywhere you go around Bhutan you'll see these giant penises uh just inscribed on the
- 3:58:03houses and stuff and this is this was Drupukunli but like I say he's the ultimate it's either he's
- 3:58:09either the the kind of the most beautiful con man or not because like i say he's he he's basically
- 3:58:19allowed to be a degenerate but if you judge him then you're not on the path of self-actualization
- 3:58:26that guy's got a sweet setup or had a sweet setup dude man he figured it out look him up anyway um
- 3:58:33try let's do this man thanks for that been a pleasure some conversation yeah i loved this
- 3:58:38love this i i don't know whether i'm any wiser for it but uh i certainly will wake up tomorrow
- 3:58:44a different person i think the same is for uh for me as well and i really appreciate it this
- 3:58:48is always a blast great let's do it again another time absolutely
- 3:58:51right you made it to the end of the show so i'm gonna go ahead and assume that you got some value
- 3:59:01out of this episode if you want to help the show please follow or subscribe right now
- 3:59:07It takes only two seconds and it massively helps.
- 3:59:12If you've got another 30 seconds, well, leave a review on Apple Podcasts or Spotify.
- 3:59:18Honest reviews apparently are the single best way to help new listeners find us.
- 3:59:23If you want more, you can get bonus episodes and early access to episodes on fountain.fm
- 3:59:30for the price of one coffee for a whole month.
- 3:59:34or you can support the mission by streaming sats on any of the value for value platforms
- 3:59:39or use the bitcoin and fiat donation links in the show notes every contribution to the show
- 3:59:46makes a real difference believe me and one more thing if you know someone who'd love this episode
- 3:59:51share it directly with them that personal share beats every algorithm in the book all the links
- 3:59:57the community and the newsletter are in the show notes and jump on over to freecities.fm
- 4:00:04for a deep dive into all the episodes with descriptions and more that's the new free
- 4:00:11podcast website freecities.fm have a good week and see you next time