Latest / In the Loop AI / AI Robots and The $10T Arms Race

Transcript

AI Robots and The $10T Arms Race · In the Loop AI

0:03
David: Welcome to the AI Roll-Up brought to you by the Limitless podcast,
0:06
David: where we say up to speed with the emerging trends and developments in the AI space.
0:10
David: I'm David Hoffman here with my two co-hosts Ajaz and Josh. Ajaz, how are you doing?
0:15
Ejaaz: I'm doing well, David. It's been quite the week. Less model,
0:20
Ejaaz: frontier model breakthroughs this week.
0:22
David: No frontier model breakthroughs this week? I think that's a first.
0:25
Ejaaz: There's a few, but we're putting them on the back burner because there's some
0:28
Ejaaz: more important things to talk about.
0:30
Ejaaz: It's been an M&A week. So, you know, billions of dollars are being spent to
0:34
Ejaaz: purchase or acquire some of these AI startups that we've spoken about previously on the show.
0:38
Ejaaz: But I think more importantly, I think AI's looked like kind of like this gargantuous
0:44
Ejaaz: cloud, right? And it's kind of ominous.
0:47
Ejaaz: I'm kind of like, is this cloud going to rain on me or is it just going to pass
0:50
Ejaaz: by peacefully and just like kind of like be nice to look at?
0:53
Ejaaz: And I'm starting to see the light. I'm starting to see the kind of formation
0:56
Ejaaz: of what this AI stuff is going to look like.
0:59
Ejaaz: And the reason is, I think we've spoken about a bunch of different things,
1:02
Ejaaz: right? We've spoken about agents.
1:04
Ejaaz: We've spoken about autonomous execution, like, hey, these agents are going to
1:07
Ejaaz: go run and do a bunch of things for you, run companies for you.
1:10
Ejaaz: We've spoken about all these amazing models and how smart they are,
1:14
Ejaaz: but we haven't really pieced together what the end goal is going to look like.
1:18
Ejaaz: We're starting to see a lot more of that this week by some of the moves made
1:21
Ejaaz: by OpenAI and Anthropic. So I'm really excited to get into that.
1:24
David: Yeah, as we talk more and more about AI, I think we are all starting to synthesize
1:28
David: a little bit some of the conversations that we've had on previous episodes,
1:31
David: put them all together and kind of understand the contours and the shapes as
1:35
David: this thing becomes more and more real.
1:38
David: Josh, I introduced this as the AI Roll-Up brought to you by the Limitless podcast.
1:42
David: That is the first time that this has ever been uttered on this podcast feed,
1:46
David: which is the Bankless podcast feed.
1:48
David: But we are launching Limitless this week, which is very exciting.
1:51
David: Talk to me, Josh, about Limitless.
1:54
Josh: Oh man, big week. Limitless kind of stemmed from like a natural extension of
1:58
Josh: our curiosity at Bankless.
1:59
Josh: Bankless is mostly a crypto company. We started getting interested in AI and
2:02
Josh: then naturally we kind of spread out to robotics, manufacturing and energy and
2:06
Josh: all this weird stuff that's happening as downstream effects of this new wave of AI.
2:10
Josh: So Limitless is a place for that. It's the home for all of the stuff that is
2:14
Josh: not crypto, is not economic, is not macro related. It's all the frontier technology, the AI stuff.
2:20
Josh: That's where this show is actually going to be Living is on the Limitless feed.
2:23
Josh: So if you're interested in tech, if you're interested in frontier technology
2:26
Josh: of any sorts, if you're interested in AI, Limitless will be the new home base
2:30
Josh: for all of that. And we're launching this week.
2:33
David: Yeah, yeah. So that is the call to action. What we are doing is we are taking
2:37
David: all of the AI content, all the frontier tech content, which we've been pretty
2:41
David: aggressively exploring at Bankless.
2:42
David: We're putting it into its own feed, which it totally deserves.
2:46
David: And I know there's a lot of people who have been following Bankless for the
2:48
David: AI stuff, but not for the crypto stuff.
2:51
David: And so we are creating a dedicated feed for that. There's also followers of
2:53
David: Bankless, listeners of Bankless who are like, another AI episode.
2:56
David: Weird reaction, but I get it.
2:59
David: It's interesting, like me and Jaws and Joshua, we're all here in New York,
3:02
David: and we'll go to crypto events that are about AI.
3:06
David: And so it's actually a crypto event by crypto people for crypto people,
3:09
David: but it's about the AI subject. So AI has broad appeal. So we're giving it its
3:13
David: own dedicated podcast feed. You're still going to be hearing these episodes
3:16
David: on the Bankless feed as we incubate the new Limitless podcast.
3:20
David: So there is a link in the show notes to go subscribe to the podcast to get that
3:23
David: podcast in your podcast player and also subscribe to the YouTube because we
3:28
David: are making some AI first, AI primary content there.
3:31
David: And then we're also going to do the regular bankless thing of interviews with
3:34
David: big founders, big investors, entrepreneurs in this space who are really helping shape this future.
3:40
David: And also just to double down on what Josh was saying, AI is definitely,
3:44
David: but also frontier tech. So we're talking like rockets. We're talking robotics.
3:47
David: We're talking everything that's going to make the future even weirder than it
3:50
David: already is because it feels like that is coming very, very soon.
3:53
David: But this is the AI roll-up, which again, also happens every single week.
3:57
David: Topics of the week that we're going to talk about, Ijah has already hinted it.
4:00
David: OpenAI making big moves, a $3 billion arms race to gobble up all of the valuable
4:04
David: startups in the AI space. Visa is giving AI agents credit cards,
4:07
David: letting them go wild on the internet.
4:09
David: What happens when you put an LLM into a robot, a real physical robot?
4:14
David: What happens next? And lastly, the hard problem of interpretability.
4:17
David: Why it's important for making sure we can save the future of humanity by cracking
4:22
David: this interpretability problem.
4:24
David: But Jaws, we are going to start with OpenAI making three big,
4:27
David: big moves, a $3 billion arms race. What happened in OpenAI news this week?
4:31
Ejaaz: Okay, so let's get into it. This has been basically the headline news all week
4:35
Ejaaz: and everyone just couldn't stop talking about it yesterday.
4:39
Ejaaz: OpenAI is officially making its first major acquisition by purchasing a company
4:44
Ejaaz: called Windsurf for $3 billion with a B.
4:50
Ejaaz: Now, if you're wondering what on earth Windsurf is, you might have heard David,
4:54
Ejaaz: Josh, and I speak about a company previously on this series called Cursor.
4:59
Ejaaz: And Cursor is basically this kind of vibe coding platform where you can type
5:04
Ejaaz: in a prompt, typically like you do on ChatGPT, but instead you could ask it
5:08
Ejaaz: to create a new app for you or a new piece of software, or maybe even a fun
5:11
Ejaaz: game that you know and love.
5:13
Ejaaz: And it would just code it all right up in front of you.
5:16
Ejaaz: And let's say the environment or the software that allows you to do that is
5:22
Ejaaz: something called an IDE, which stands for Integrated Development Environment.
5:27
Ejaaz: If you're wondering what that
5:27
Ejaaz: is, think of it as a software suite which includes like a coding editor.
5:34
Ejaaz: A compiler and a bunch of other things that you need to basically make software
5:38
Ejaaz: it just makes it really really easy for you and then it adds ai on top of it
5:42
Ejaaz: so you just speak to it like a normal person and it just kind of like pops up an app now,
5:46
Ejaaz: windsurf is in the top two of these types of companies you've got cursor then
5:51
Ejaaz: you pretty much have windsurf at this point,
5:53
Ejaaz: and open ai will now compete directly with things like Cursor,
5:58
Ejaaz: Lovable, and Replit, which do very similar things.
6:01
Ejaaz: And in my opinion, this confirms that there's pretty much an arms race to build
6:05
Ejaaz: the best AI coding assistants ever. So imagine if you could be like the number
6:10
Ejaaz: one tool or platform to the leading software engineers and companies of tomorrow.
6:15
Ejaaz: That's basically the TAM, the Total Adjustable Market. That's basically the
6:19
Ejaaz: market that you're going after.
6:20
Ejaaz: Now, one kind of like interesting caveat that I want to include here is that,
6:26
Ejaaz: This happened after OpenAI tried to buy Cursor. That's the number one platform that does this twice.
6:32
David: So they had to get the second best.
6:34
Ejaaz: Yeah, they basically had to get the second best. And kind of like my initial
6:38
Ejaaz: gut reaction when I read that was,
6:40
Ejaaz: I wonder if Cursor is either holding out for a Google or a Meta to purchase
6:45
Ejaaz: them and go for an even bigger number because they don't want to be bought by Sam Altman's startup.
6:51
Ejaaz: But it's a curious move. and a few other things i want to highlight is this
6:56
Ejaaz: acquisition follows um some major moves by other companies that are doing something
7:01
Ejaaz: similar so google for example launched their own cursor competitor called firebase
7:06
Ejaaz: and if you pull up this tweet david it's actually pretty good and i think this
7:09
Ejaaz: um this example will be quite a good visual example so people can understand
7:13
Ejaaz: you know what on earth this does
7:14
Ejaaz: so what you're watching right now um for those of you who are just listening
7:17
Ejaaz: is a video of someone just sketching out a very, it looks like a three-year-old's drawn it,
7:23
Ejaaz: kind of layout of a website, of an app that can basically help someone draw.
7:28
Ejaaz: And then what you see on the right side is the Google's Firebase platform creating it.
7:34
Ejaaz: And you can actually interact with the tool and draw and paint just like you
7:37
Ejaaz: did on like Microsoft Paint.
7:39
Ejaaz: Very simple, basic example. But can you imagine this being extended out to something else?
7:43
Ejaaz: It's pretty insane. What's your take on this, David?
7:45
David: I think for the listeners, just imagine Microsoft Paint, but on the other side
7:50
David: of it, outcomes and an application, a functioning application.
7:54
David: So, you know, Microsoft Paint on one side, functioning application on the other.
7:58
David: I'm actually curious to hear Josh's take on what's going on here.
8:01
Josh: I think it's cool. I think the second I saw this news, it reminded me of news
8:05
Josh: from a few weeks ago, which was Google acquiring Wiz for $32 billion, I think it was.
8:10
Josh: I think this is an attempt at OpenAI to kind of make their way up the stack.
8:14
Josh: And it's not just about serving their own stack better. It's about now actually
8:17
Josh: embedding themselves in the decision-making layers above that stack so they
8:21
Josh: could shape the flow of traffic regardless of the endpoint.
8:24
Josh: In the case of Wiz, Wiz was kind of like a security aggregator company.
8:27
Josh: So Google owns Google Cloud, but it doesn't have any control over AWS or Microsoft's Azure servers.
8:34
Josh: So what acquiring Wiz does is it allows them, it gives them upstream access
8:39
Josh: to that, like security telemetry, the customer data flows, the threat intelligence,
8:42
Josh: a lot of the data that they wouldn't otherwise have.
8:44
Josh: And I think that's kind of what OpenAI is doing here with Windsurf,
8:47
Josh: where ChatGPT clearly dominates the consumer product space.
8:51
Josh: They own that. They do very well. But in enterprises, a lot of companies are
8:55
Josh: looking for a thing that people are calling model orchestration,
8:58
Josh: which is kind of like, which model will serve my use case the best?
9:01
Josh: And they're looking for aggregators to decide and make that decision for them.
9:05
Josh: Well, OpenAI just bought that aggregator. And they now own this routing and
9:09
Josh: orchestration layer that acts like across this load balancing metric.
9:13
Josh: So now, strategically, they get visibility over influence, over who's using
9:17
Josh: what, what they're using for. And they can optimize their models for these use
9:20
Josh: cases because they have the data.
9:22
Josh: So now going forward, you might see ChatGPT starts to get chosen a lot more
9:26
Josh: by these aggregators because it has all the data of why it's choosing another
9:30
Josh: model that isn't theirs. And I think that's like a pretty interesting play.
9:33
David: Is there a parallel here between like there are deals that Google has made in
9:37
David: the past with browsers about like, hey, we will pay you X amount of money to
9:41
David: make us the default search engine in your browser. So when you type a question
9:44
David: to the toolbar, it automatically routes to Google.
9:48
David: It makes a Google query and Google gets the data from that and the information
9:51
David: for that to make their product better.
9:52
David: It sounds like this is something similar where they are just acquiring this
9:56
David: endpoint, which has a direct relationship with its users. So Windsor has ownership
10:03
David: over the users. has the user relationship. But in the back of Windsurf, there's ChatGPT.
10:07
David: So ChatGPT now just acquires that user relationship. It gets ChatGPT to be used by Windsurf.
10:14
David: And so there's just, it's just more usage for ChatGPT. And then there's also
10:17
David: information about what ChatGPT, how it needs to adapt the information,
10:21
David: the data, like you said, it goes back into ChatGPT.
10:23
David: Is there a parallel there or am I hallucinating here?
10:26
Josh: Probably. It feels like there's a fine line where people go to Windsurf as Windsurf customers.
10:30
Josh: They don't want to be spoon-fed ChatGPT if it's not the ideal model,
10:34
Josh: so that could hurt with customers.
10:35
Josh: I assume they probably won't do it that explicitly, but the implicit thing that
10:38
Josh: they could extract value from is understanding the user's needs and the wants
10:42
Josh: and when certain models are chosen and kind of optimize future versions of the ChatGPT
10:48
Josh: language model to serve those people.
10:50
Josh: So the ideal case is that ChatGPT will actually be able to serve all of these
10:55
Josh: while still gaining data from what people like about the other models.
11:00
Ejaaz: You could also argue that it's just, to your point, Josh, data that they want.
11:05
Ejaaz: Data is the end game, right?
11:07
Ejaaz: In whatever shape or form it comes in, in whatever kind of outlet that you can
11:11
Ejaaz: extract it from, that's what they want. If they want developer activity,
11:14
Ejaaz: if they want to grab it at the consumer level, they already have it at the end
11:18
Ejaaz: consumer level. So now they want to try and get like all the developer activity as much as they can.
11:23
Ejaaz: I wonder what type of a model or even end-to-end experience you can create based off of that, right?
11:29
Ejaaz: And to kind of like strengthen your argument here, I noticed that Google launched
11:34
Ejaaz: something associated with integrated computer use this week as well.
11:40
Ejaaz: They have something called AI Studio, which is kind of like this environment
11:44
Ejaaz: that you're describing, Josh, which doesn't just include kind of like the end
11:48
Ejaaz: user stuff, which OpenAI dominates on, but it also allows developers to kind
11:53
Ejaaz: of like build this kind of like,
11:55
Ejaaz: synonymous end-to-end app experience, right?
11:58
Ejaaz: And what I noticed about this integrated computer use thing is that,
12:01
Ejaaz: you know, they really want to just own end-to-end product development, the entire pipeline.
12:07
Ejaaz: And if you think about it, like you allow them to use their no-code prompt-friendly
12:13
Ejaaz: platform to build your product, right? Leveraging their models and you can inference
12:17
Ejaaz: it through their like cloud system.
12:19
Ejaaz: And then on top of that, you can allow anyone to use it for themselves involves
12:23
Ejaaz: via computer use so if you're an end user right and you're saying hey i really
12:26
Ejaaz: like that app that this developer's built well you can run it locally on your,
12:31
Ejaaz: computer now and you can inference any model that you want say like hey i don't
12:34
Ejaaz: like that it's using this model can we use something else sure like google's
12:38
Ejaaz: ai studio solves for that and i think that's like a really important nuance
12:42
Ejaaz: that's going to make it super super sticky and how they personalize this,
12:46
Ejaaz: experience for users i don't actually quite know i wonder if you have any ideas
12:50
Ejaaz: but um just a very subtle but important improvement i think is is worth noting.
12:55
David: I've brought this metaphor up before in the past on this show but uh the whole
12:59
David: entire um vertical the gaming vertical of uh like league of legends and dota
13:04
David: is like arena battlers or whatever uh
13:07
David: billion dollar industry in just that one type of gaming structure and that gaming
13:12
David: structure was created by a
13:14
David: starcraft mod so some individual using the starcraft map editor and like map
13:20
David: editing engine created uh
13:23
David: this like mod of starcraft that created this structure that basically is is
13:27
David: what uh the moba arena battler league of legends
13:30
Josh: Multiplayer online battle arena
13:32
David: Thank you i don't play league of legends But a
13:38
David: multi-billion dollar industry got created because of this modding ability,
13:42
David: because of the modding ecosystem around this one base game, which was very,
13:46
David: very valuable called StarCraft, but now it spawned an insane amount of value.
13:51
David: And so what I'm seeing here is this is a modding engine for apps.
13:55
David: And the downstream value that could come out of this just give...
14:01
David: Give creativity, give creative tools to people who want to be creative.
14:04
David: And now it's not even developers. It's anyone who's like frustrated that they
14:08
David: are using this app, but there's this one button that's missing that they wish
14:12
David: they could press and it could do something.
14:14
David: And now all of a sudden the creation of that app, if they truly discover value
14:19
David: in it, maybe so do other people as well.
14:21
David: And they just need that one extra little module to hook it into the app.
14:24
David: And now all of a sudden that is a valuable piece of infrastructure that is free
14:28
David: to roam the internet. And so that's what I'm seeing here and why I'm pretty excited about this.
14:32
Ejaaz: You know, something you just reminded me of something, David,
14:35
Ejaaz: when you describe that kind of open modding kind of ecosystem,
14:40
Ejaaz: it's kind of what we're seeing amongst a lot of AI trends today, right?
14:45
Ejaaz: So we've spoken about reinforcement learning, which is like,
14:48
Ejaaz: you know, a very popular post and pre-training method to get AI models to become
14:53
Ejaaz: smarter. And it doesn't require a hell of a lot of compute. It just requires
14:56
Ejaaz: you to give it kind of like some reasoning logic and it gets better and better.
15:00
Ejaaz: I've noticed that the primary way that a lot of these models are learning is
15:05
Ejaaz: through open reinforcement learning gyms.
15:08
Ejaaz: So think of it as like a kind of like a Pokemon gym, right? And you could take
15:12
Ejaaz: your Pokemon, but in this case, your Pokemon is a model.
15:15
Ejaaz: You can put it in the gym and you could train it. It battles over and over again until...
15:20
Ejaaz: It gets smarter and smarter. And you can like orient it around like math,
15:23
Ejaaz: coding, whatever it might be.
15:26
Ejaaz: The thing that accelerates reinforcement learning, David, is just allowing a
15:30
Ejaaz: bunch of anyone, any humans to create their own environments.
15:34
Ejaaz: And then you could just send your model to all of them in a day,
15:37
Ejaaz: or you could just pick the best ones. And it kind of like ranks itself via an open source method.
15:42
Ejaaz: Another example is in training, right? Where previously, like it was just one
15:48
Ejaaz: specific data center, then it was like, oh shit, we need more power.
15:52
Ejaaz: Let's move all the data centers together. But like, let's keep them close because,
15:56
Ejaaz: you know, we need like a high feedback loop.
15:59
Ejaaz: And now it's getting even more distributed. And we'll talk about that later.
16:02
Ejaaz: But I'm just noticing it's like this, dare I say, open source ecosystem that
16:06
Ejaaz: is like purely benefiting the way that we advance AI right now.
16:09
David: We talked last week about the downstream implications of OpenAI's memory update
16:13
David: and how it can remember all of its chats with you.
16:17
David: This saga continues this arc continues what happened in the last seven days to Joss?
16:21
Ejaaz: Okay, so to give context on when we last mentioned this update,
16:25
Ejaaz: David, it was really positive, right?
16:27
Ejaaz: Because we were like, hey, ChatGBT is going to remember everything you've ever said to it.
16:32
David: ChatGBT is turning into a friend. It's turning into an ally.
16:35
Ejaaz: Oh, not just a friend, your best friend. Your best friend. Your second brain. Your psyche.
16:42
Ejaaz: Exactly. You're going to offload your conscience to this thing,
16:45
Ejaaz: basically, and you're going to love it, right?
16:47
Ejaaz: And there were a lot of, like, you know, kind of ominous implications from that.
16:51
Ejaaz: But overall, it was like net net good because the more personalized your AI
16:54
Ejaaz: experience is, you could argue the better and more sticky the product is going to be.
16:58
Ejaaz: Now, there's a dark side to this. And OpenAI, funnily enough,
17:02
Ejaaz: didn't really announce this on any major headlines, which is they updated your
17:06
Ejaaz: memory in a much more creepier way,
17:09
Ejaaz: such that in whatever way you prompt in ChatGPT, you know, sometimes it does a web search for you.
17:15
Ejaaz: So let's say like, hey, can you tell me some of the hottest recipes?
17:19
Ejaaz: I don't know why I keep talking about recipes on the show, by the way.
17:21
Ejaaz: I think I'm just like hungry when I do this episode.
17:23
Ejaaz: But let's say you prompt ChatGPT and you say like, hey, like pull out the top
17:27
Ejaaz: five recipes to create roast chicken or something, right? It does a web search for you.
17:32
Ejaaz: But now with this update, ChatGPT has the permission, rather your explicit permission,
17:38
Ejaaz: to change the way you've worded your prompt.
17:41
Ejaaz: Now, the reason they're giving to you there is just so that they can make your
17:47
Ejaaz: prompt more effective. Right.
17:48
David: But I'm more attuned to what your interests are.
17:51
Ejaaz: More attuned to what, and I'm like, well, hang on a second. How do you know
17:54
Ejaaz: what I want better than I want, right? And we get into like this really weird
17:57
Ejaaz: territory where it's like, hey, can you tell me like what kind of clothes do
18:01
Ejaaz: you think might fit my vibe for this evening?
18:03
Ejaaz: And it starts, you know, pulling like ad sponsors type situation,
18:08
Ejaaz: but then it feeds it to you as if like, hey, this is what you really want,
18:10
Ejaaz: right? And I'm thinking, hang on, like, how can you like change my words?
18:14
Ejaaz: And like, am I held legally wrong?
18:17
Ejaaz: Liable to that like it gets super weird and i'm curious you know how this can
18:21
Ejaaz: get kind of like taken out of context to like some kind of black mirror episode um.
18:28
David: The pattern that i'm seeing here is like the pre-2015 era of the internet which
18:32
David: was already starting to decay at this time but um
18:34
David: you know facebook and cambridge analytica and all of that debacle what we all
18:39
David: learned was that facebook was showing conservatives one version of the internet
18:43
David: one version of the truth It was showing liberals a different version of the internet.
18:47
David: And people were all learning like, oh, my version of the internet is attuned
18:51
David: to me based off of the data that I have exposed about myself to the internet.
18:55
David: And all of a sudden, like my internet is not your internet. It all used to be
18:58
David: one internet. We all looked at the internet and we saw a single source of truth
19:02
David: and the algorithms hadn't corrupted that.
19:04
David: And we're all looking at the same facts. We're all looking at the same news articles.
19:08
David: But as algorithms became more precise about what their goals were,
19:12
David: like, oh, we can get this user to stay online on our platform more if we feed
19:17
David: them content that is more attuned to what their beliefs, then then we're going to do that.
19:20
David: And all of a sudden, you know, everyone got their own interpretation of the
19:23
David: Internet. And we also started to like society started to split into factions.
19:26
David: Right. The right got more right. The left got more left.
19:29
David: And it's because of this curation, which we all want. I want ChatGPT to curate
19:34
David: the best for me, the best for my my likes.
19:37
David: But this is the same pattern that I'm seeing Is like ChatGPT is now Profiling me
19:43
David: And judging me And putting me into a box That I'm not necessarily aware That
19:48
David: it is the box that it is putting me in But nonetheless that is the box that
19:52
David: I'm going into And all of a sudden I will be in a box That
19:55
David: Other the rest of society will not be in and they will be in their own box and
19:58
David: now all of a sudden we are not again
20:00
David: connected as this as a species we are now like profiled and segregated a base
20:05
David: of our interests in our and how chat gpt is being tuned so that is what is triggering
20:09
David: in in me in this hearing this
20:12
Josh: It's this uh this like increasing hyper personalization that we're seeing there's
20:15
Josh: two trends which is the hyper personalization things are created specifically for you
20:19
Josh: and also the relinquishing of all your privacy and data for the better experience
20:25
Josh: which is just a trend that i do not see stopping anytime soon and the
20:29
Josh: results are probably a more fractured idea space because things do become personalized
20:34
Josh: to you and you're seeing this different reality than everybody else
20:37
Josh: um the question is is that better and and what is the incentive from the person
20:41
Josh: who's serving it to you is the incentive to get you to maximize your time spent on the service
20:46
Josh: or is it to give you a truthful answer so you can leave like i love the example
20:50
Josh: you used last week david where you said you use like five or ten percent of
20:53
Josh: your screen time on chat gpt and it's the most valuable time that you spend
20:56
Josh: is that gonna be the outcome where you come you get served and you leave or
21:01
Josh: is it an attempt to just trap us in ecosystem lock maximally extract uh that type of thing and that's
21:08
Josh: it's definitely a step towards that direction
21:09
David: Yeah i definitely see that as an emerging theme of these conversations the ai
21:13
David: roll up the things that the news that we're processing on this base there's
21:16
David: like the dark path and the light path there's like the utopia and the dystopia
21:20
David: and there's the utopia where we want where this
21:23
David: chat gpt intelligence is minimally invasive and maximally value add to our lives
21:28
David: and allows us to be human connect with each other better be informed be more
21:32
David: intelligent and that's the happy path and then there's the dark path of
21:35
David: it encourages brain rot But it encourages isolation.
21:39
David: You guys see like the meta Zuckerberg announcements of like,
21:42
David: oh, yeah, we're actually just going to make online AIs and you're going to be
21:47
David: friends with them. And so like the original Facebook notion was like,
21:51
David: we're going to make everyone connected. We're going to connect everyone.
21:53
David: And now it's now the modern day mission of Facebook, implicit mission of Facebook
21:58
David: is we're going to isolate everyone.
22:00
David: And, you know, the modern day, the happy path of ChatGPT is we're going to make
22:05
David: everyone superpowers. We're going to give everyone godlike intelligence.
22:08
David: And then the unhappy path is we're going to do brain rot even faster, even better this time.
22:15
David: I'm worried that the incentives just point towards the brain rot all the time.
22:21
Ejaaz: Well, I mean, what's the saying around show me who makes the money and I'll
22:24
Ejaaz: show you the incentives or maybe it's like the other way around.
22:27
Josh: It's like, let's show me the incentive, I'll show you the outcome.
22:30
Ejaaz: Exactly right. So let's track it all the way up to the top.
22:34
Ejaaz: Okay, OpenAI has an amazing product and they state all these different things
22:38
Ejaaz: like, hey, we want to help you out. We want to make you a better person,
22:41
Ejaaz: a better learner, blah, blah, blah, blah, blah.
22:43
Ejaaz: But, you know, to your point, like they want retention. They want max usage
22:47
Ejaaz: 24-7 of people on this thing.
22:51
Ejaaz: And where I get even more worried or concerned is in the case of like Facebook, Snapchat, Instagram.
23:01
Ejaaz: You can take the phone away, right? You can put it away.
23:04
Ejaaz: You can stop. There's only so many pictures you can post in a day.
23:07
Ejaaz: And then you're like, okay, right, I'll put it down.
23:10
Ejaaz: But when it becomes your entire life, your thinking vehicle,
23:13
Ejaaz: and in some cases, your personality.
23:16
Ejaaz: I saw the most random thing of like, I think we mentioned it last week of PhD
23:20
Ejaaz: students using chat GPT prompts to pick up dates and stuff.
23:25
Ejaaz: And I'm just like, well, hang on a second. Like, now it's getting involved in your actions.
23:29
Ejaaz: It's going to tell me what to buy. It's going to tell me how to look.
23:31
Ejaaz: And I'm like, that is an incredibly more stickier function than like Facebook
23:36
Ejaaz: trying to influence your opinion on certain things.
23:38
Ejaaz: So I really hope that it is not going to become the darker outcome,
23:43
Ejaaz: but I can't see a path where it doesn't unless
23:47
Ejaaz: for some reason shareholders of these, you know, corporate companies are kind
23:50
Ejaaz: of like, oh, yeah, you know what, we don't want to fuck humans up too much or like too bad.
23:56
Ejaaz: But actually quite interesting on that case, guys, did you see the news around,
24:02
Ejaaz: open AI's, well, I don't wanna say restructuring, but rather structuring,
24:07
Ejaaz: which kind of like confirmed what they were before. Did you guys see that?
24:12
David: I'm going to need to be informed here.
24:14
Ejaaz: Yeah. Okay. So basically, OpenAI, I'll actually take this direct quote from Sam on his blog post.
24:22
Ejaaz: OpenAI was founded as a nonprofit.
24:25
Ejaaz: Is today a nonprofit that oversees and controls the for-profit?
24:30
Ejaaz: And going forward, will remain a nonprofit that oversees and controls the for-profit?
24:35
Ejaaz: That will not change. Now, what he's referencing here is a long-lasting,
24:41
Ejaaz: highly publicized debate or debacle, rather, that OpenAI was founded as a nonprofit.
24:47
Ejaaz: And Sam was all like, this is for the good of AI. This is back in the day when
24:51
Ejaaz: both Elon Musk and Sam, and if you guys didn't know this, had founded OpenAI.
24:56
Ejaaz: And they were working on creating AI for the betterment of humanity and the people.
25:01
Ejaaz: But at some point, Sam decided, you know what?
25:04
Ejaaz: I kind of want to make this a for profit because there's a lot of money to be made in this AI thing.
25:09
David: Well, also, he needed to to keep opening AI alive because you need the profit
25:14
David: incentive to attract the best talent, to attract investors, to stay ahead of the game.
25:18
David: And so, yes, it was a trap. It always needed to be like this.
25:21
David: As soon as there was any amount of legitimate competition, the nonprofit model was a failure scenario.
25:26
Josh: Yes, primarily to raise money for training. because in order to do those big
25:29
Josh: training runs, you needed money. And in order to attract investors,
25:32
Josh: you needed to have some sort of promise to give them money back.
25:35
Ejaaz: Right, exactly. So that is like, you know, the greenfield really good way to look at this.
25:40
Ejaaz: And I think like arguably to this extent, that's what Sam's been doing.
25:45
Ejaaz: You know, he set up, what's it called? Stargate or whatever it's called,
25:48
Ejaaz: like the biggest basically data center factory that is being built,
25:51
Ejaaz: I think in Texas or somewhere like that.
25:54
Ejaaz: And he's investing the funds that he's raised in exactly where he said he was going to do it.
25:59
Ejaaz: But at what point does the full profit then become maybe a bit of an issue?
26:04
Ejaaz: Now, I personally don't really see an issue in this turning into a full profit.
26:07
Ejaaz: In fact, I think it actually should be a full profit.
26:09
Ejaaz: If I'm going to be using this product, I want it to eventually get better.
26:13
Ejaaz: I don't want it to become some kind of,
26:14
Ejaaz: communist table stake to ai kind of thing i'm just gonna i'm just gonna go to
26:18
Ejaaz: the product that works better um so i don't really get why there's been so much
26:22
Ejaaz: kind of backlash against this i think it's because sam had,
26:26
Ejaaz: quite the lead with open air and now it's become kind of like table stakes and
26:30
Ejaaz: maybe this is a bit of a competitive move from you know uh,
26:34
Ejaaz: x and elon musk and open ai and stuff who have been filing a lawsuit against
26:37
Ejaaz: him but really interesting to see that like open airs kind of back down and said okay okay okay
26:42
Ejaaz: we'll keep it we'll keep the non-profit in control of the for-profit and wait
26:46
Ejaaz: until everything everyone's calmed down and then we'll kind of maybe flick it back to a for-profit.
26:52
David: I think ultimately that all of this is going to turn into a nothing burger the
26:55
David: non-profit wrapper around a for-profit company is just the same thing as a board
26:59
David: of directors around a for-profit company it's the same structure it's just a
27:03
David: for-profit company with extra steps
27:05
David: uh we've already figured out like general c-corp incentive structures there's
27:09
David: a reason why the delaware c-corp is what it is it is a science that has been well attuned.
27:14
David: We don't need to reinvent the wheel here. I don't know why.
27:17
David: I don't know the story behind the nonprofit genesis of OpenAI.
27:21
David: Maybe it was just an anomaly.
27:23
David: Ultimately, as time progresses, I think this is just going to be this weird
27:26
David: anomaly about how OpenAI got started.
27:28
David: And it's going to ultimately look like a for-profit company,
27:31
David: like all the rest, kind of like how it already looks. I don't know.
27:34
David: Josh, that's my take. What do you think?
27:35
Josh: Yeah. The inception of OpenAI and the reason it's called OpenAI is because at
27:38
Josh: the time Google had invented the transformer, they were becoming a powerhouse in the world of AI.
27:42
Josh: And Elon and Sam had this vision that they wanted a counterbalance to a monopolistic
27:48
Josh: and super intelligence and the counterbalance was an open source to AI that's
27:51
Josh: for the people. So that was the idea on paper.
27:54
Josh: In practice, you need a lot of money to buy these training models.
27:58
Josh: You need to buy the GPUs to train the models.
28:00
Josh: You need a lot of funding. There are profit motives that are required in order
28:04
Josh: to create consumer products that get this product out there.
28:06
Josh: So it got kind of clouded over time. And I think if...
28:09
Josh: They had just started as a for-profit organization. None of this would have made a difference.
28:13
Josh: The thing that I'm really interested in, less than the company structure,
28:15
Josh: because it feels kind of irrelevant. They are for-profit.
28:18
Josh: They will hopefully send some gimmies out, is where they stand on the open part of it.
28:24
Josh: Where will the company state, like, hey, we are going to release these models
28:28
Josh: to the public as a public good service, and we're going to keep these models closed?
28:31
Josh: I'm more interested in that internally, is how they're thinking about releasing
28:34
Josh: those models to the public and actually maintaining the open sense,
28:37
Josh: that initial mission of distributing the computes and the intelligence.
28:42
David: We've talked in the past on this podcast about how all of the frontier models
28:47
David: just become owned by the value of these models, become owned by the public domain
28:53
David: pretty quickly because of the rat race of open source.
28:57
David: Like open source is, I don't know, six months behind, nine months behind the
29:01
David: frontier models. It is not more than a year behind.
29:04
David: And so the value of the frontier models, whoever's got the best frontier model,
29:08
David: that value will show up in the open source domain within 12 months.
29:12
David: And so I'm not sure what the value is. Like, does it matter if OpenAI keeps
29:17
David: their models behind a silo, a closed silo?
29:22
David: Maybe they're a hypocrite, but I'm not sure that matters because ultimately
29:24
David: the open source arena always catches up to the value of the best model that
29:29
David: we've produced somewhere on the
29:30
David: earth. And that is inclusive of all the models coming out of China too.
29:34
Josh: This has been true in the past. I think the scales are a bit different.
29:39
Josh: I actually saw this interesting post by Kevin Whale this morning.
29:41
Josh: He's like the chief product officer of OpenAI. And it was about a blog post
29:44
Josh: where other countries are actually
29:46
Josh: reaching out to OpenAI to build Project Stargates in their own country.
29:50
Josh: And that diminishes the free marketization of these large language models where
29:55
Josh: now there's not a free market competing for the best spot and releasing it.
29:58
Josh: There is a country that wants to consult with a single company that wants to
30:03
Josh: integrate their AI into the way that they populate AI through their country.
30:07
Josh: So it feels like the scales are increasing and the breadth of this is narrowing
30:13
Josh: in the sense that they're looking for a source of truth from one person instead
30:18
Josh: of this open source version.
30:20
Josh: And that feels like where things can get a little shaky.
30:23
David: Interesting. Okay, so what you're saying is that the assumption that there's
30:26
David: compression amongst all the frontier models, there's like seven,
30:29
David: eight frontier models out there, all these different AI labs,
30:32
David: many of them, all ruthlessly competing.
30:35
David: And because of that competition, there's a great equalization because the learnings
30:39
David: of one frontier model become the learnings of the other, and that value gets passed around.
30:43
David: And then because it's just passed around, it becomes available in the open source domain.
30:47
David: What you're saying is that that assumption that that is going to be the way
30:50
David: that it is is not a perfect assumption and we should be wary that actually there
30:54
David: are very strong economies of scale here and the possibility of one or two or
31:00
David: a very low number of frontier models
31:02
David: make a break for it and actually become very large and have economies of scale
31:07
David: is a possibility that we should watch out for that's that's what i just heard from you
31:11
Josh: Yeah that that feels about right and um
31:14
Josh: Yeah, I think as things kind of accelerate faster, if you ask the OpenAI team
31:18
Josh: from a decade ago, when they were first getting started, that there would be,
31:22
Josh: if you told them that there would be seven or eight giant frontier model companies
31:24
Josh: that are all competing, that'd be a great thing.
31:26
Josh: So, so far we're good because the initial use case was to fight Google,
31:30
Josh: which was the single entity. So, the fact that there's seven or eight now is good.
31:33
Josh: I think it's just important to watch the narrowing of that as the velocity of
31:37
Josh: these models increases to just kind of keep checks and make sure that some aren't
31:42
Josh: really running away faster than the others.
31:45
Ejaaz: Where do you think the stickiness forms josh i i know we've discussed this previously
31:51
Ejaaz: but i think it it comes down to who creates the best,
31:56
Ejaaz: end user app whether that is some kind of developer platform or a chat gpt interface
32:02
Ejaaz: that does a bunch of stuff for you i don't really see how it could be anything
32:06
Ejaaz: to do with some kind of pioneering model unless it's like really that much better Right.
32:11
Ejaaz: All the model updates that we see, you know, over the last couple of weeks with
32:15
Ejaaz: Quan, with Gemini 2.5, Flash from Google, they all beat certain benchmarks.
32:22
Ejaaz: But I think no one really sees what that looks like until it's like in practice. Right.
32:28
Ejaaz: Unless you're like a crazy like kind of coder or whatever. So I think it comes
32:32
Ejaaz: down to like stickiness.
32:34
Ejaaz: And so if you have, and I want to hear your opinion on this,
32:38
Ejaaz: OpenAI going to other countries and,
32:41
Ejaaz: you know, committing basically a ton of data and compute to,
32:46
Ejaaz: you know, help them whatever locally own whatever OpenAI's product or models are over there.
32:53
Ejaaz: How do you think that kind of like seeds their position other than just like the model and compute?
32:57
Ejaaz: Or do you think that is enough of a moat, right? Just because like they're there.
33:02
Josh: Yeah, there's probably different layers to it. There's like the consumer and
33:04
Josh: business layer, which is
33:07
Josh: the app moat that we discussed with the data mode with memory and that's super
33:10
Josh: powerful but i think above that is the like nation-state policy level influence
33:16
Josh: where i mean we made a joke of it with the tariffs that they were generated by chat gpt but by
33:21
Josh: by using these systems for for things that are greater than just consumer applications
33:26
Josh: for making like large decisions for implementing policy for implement like think
33:30
Josh: of china for how they should
33:32
David: We tariff china or not
33:33
Josh: Yeah manage or even as as the chinese like as a Chinese citizen,
33:37
Josh: how the government should command its citizens.
33:41
Josh: I think there is like this higher level influence of AI that is more idea driven
33:46
Josh: instead of consumer driven in terms of like policy and how to run countries.
33:51
Josh: That's the part that feels, yes, more, more vibes and more heavily influenced.
33:55
Josh: So when I read this early this morning, the open eyes can be working with countries
33:58
Josh: instead of companies that was like, oh, OK, this is this. The scales are now
34:03
Josh: like getting grander and grander. Right. Right.
34:06
David: Yeah. The idea of China using AI to govern its population is not a new idea.
34:13
David: I think even Peter Thiel talked about this forever ago, which he talked about
34:17
David: like AI is highly centralizing, crypto is highly decentralizing,
34:20
David: and both of these like verticals are growing antagonistically towards each other.
34:25
David: Anyways, I think we're ready to move on from OpenAI. Let's get into the subject
34:28
David: of Visa because Visa is giving AI credit cards, and I don't know what that means. Josh, what's going on?
34:35
Ejaaz: Listen, please.
34:37
Ejaaz: Yeah. So in an unexpected move, honestly, Visa announced something called intelligent commerce,
34:45
Ejaaz: which is basically giving a bunch of autonomous, well, kind of semi-autonomous
34:50
Ejaaz: AI agents a credit card or the equivalent of like a bank account or a wallet.
34:55
Ejaaz: And it makes sense because I think AI commerce is becoming more of a thing.
34:59
Ejaaz: And I'll get into what exactly that means.
35:01
Ejaaz: But basically, the initial use case for this is that you have like an agent
35:07
Ejaaz: that you can talk to that you can use to do all the kind of like boring things
35:12
Ejaaz: that you kind of don't want to be getting to, but you need to as like daily errands, right?
35:16
Ejaaz: So maybe, hey, can you go order the weekly groceries off of Amazon for Whole
35:22
Ejaaz: Foods or whatever? Or can you organize a travel itinerary for my business trip?
35:26
Ejaaz: Or, you know, log in through my company server and do A, B, and C?
35:29
Ejaaz: Or restaurants, hotel bookings, stuff that we've spoken or discussed about on this show before.
35:36
Ejaaz: But I think the specifics is where it gets kind of interesting.
35:38
Ejaaz: So humans will basically be able to have full control over the rules and limits
35:43
Ejaaz: that an agent can operate with, right?
35:46
Ejaaz: So you can kind of determine what its spending cap it's going to be,
35:50
Ejaaz: what kinds of websites it can visit, whether it needs to be an official website
35:53
Ejaaz: address or whether they can kind of like vibe on Google and pick like the top
35:57
Ejaaz: sponsored link or whatever that might go.
35:58
Ejaaz: And, you know, that could go wrong in many different ways.
36:02
Ejaaz: What I found really interesting here in the spec is they're using tokenized
36:06
Ejaaz: digital credentials linked to the human owner.
36:09
Ejaaz: And that sounds super vague and boring, but I think is actually really important because
36:14
Ejaaz: if you think about this AI becoming eventually an extension of yourself,
36:18
Ejaaz: right, a digital identity, We've spoken about this a lot on the crypto and Web3
36:23
Ejaaz: side, which is like, oh, we should decentralize identity and it'll give you like this, you know...
36:29
Ejaaz: Self-owned financial credit score and all that kind of stuff,
36:32
Ejaaz: we're seeing Visa basically take the steps towards defining what those credentials
36:36
Ejaaz: look like for you from a financial sense, right?
36:40
Ejaaz: In addition to the spec, you can do things like dispute handling in real time
36:44
Ejaaz: because Visa has a customer support line and it's managed by Visa.
36:49
Ejaaz: And they're launching with some really cool partners, which I think is one of
36:52
Ejaaz: the most important things when you're launching a product like this.
36:55
Ejaaz: You need distribution and partners.
36:58
Ejaaz: And the partners that they're launching with are all the big dogs,
37:00
Ejaaz: like OpenAI, IBM, Anthropic, Microsoft, and Stripe.
37:05
Ejaaz: And the reason why I find this such a compelling thing, and maybe this is because
37:09
Ejaaz: I've spent a lot of time on the Web3 stuff, is that was mainly the pitch that
37:13
Ejaaz: was given for a lot of crypto AI agents, right?
37:17
Ejaaz: You know, we've spoken about this a lot, David and Josh, where we're like,
37:20
Ejaaz: okay, I think crypto is going to be pretty much the financial rails for the
37:24
Ejaaz: future of AI. And the reason why it's going to.
37:26
David: Be the case is- AI guys can't have bank accounts, but they can have Ethereum addresses.
37:29
Ejaaz: They can have Ethereum addresses.
37:30
David: I can go find 20 Ryan Shaddam tweets like that.
37:33
Ejaaz: And listen, David, I can never take down your Ethereum wallet,
37:36
Ejaaz: right? You know, you own your keys, blah, blah, blah. It's cheaper in some cases
37:40
Ejaaz: if you're using L2s or whatever.
37:42
Ejaaz: And, you know, you're able to basically access any app that anyone creates on
37:46
Ejaaz: this infrastructure known as the blockchain.
37:48
Ejaaz: Now Visa's coming along. Oh, by the way, the antithesis for TradFi was it's too expensive.
37:54
Ejaaz: They'll never scale. Their infrastructure is too siloed, right?
37:59
Ejaaz: And now you have Visa come along being like, hey, it's not really that big a deal.
38:03
Ejaaz: We will gladly give your agents their own wallet.
38:07
Ejaaz: We'll give you even better controls over it. And we'll give you a customer support
38:11
Ejaaz: line if anything goes wrong. Can't do that with a blockchain, can you?
38:14
David: You know, in hindsight, when we are talking about a revolution of intelligence,
38:17
David: the idea that AI won't be able to learn how to use credit cards kind of seems dumb.
38:21
Ejaaz: Yeah, so I think this is interesting because this isn't an update in silo from Visa.
38:27
Ejaaz: MasterCard also had an update, funnily enough, this week. So I feel like they
38:31
Ejaaz: were playing off of each other when one announced first that they're announcing
38:36
Ejaaz: payments for agents called AgentPay.
38:39
Ejaaz: It does a lot of the same thing that I just described for Visa's new product,
38:42
Ejaaz: but specifically this takes place within conversations that users are having with AI.
38:48
Ejaaz: For example, it'll integrate directly with ChatGPT or Microsoft's AI co-pilot.
38:54
Ejaaz: It'll also leverage things like memory data from each of the platforms that
38:57
Ejaaz: you use. So for example, OpenAI's memory that it has on you,
39:01
Ejaaz: it'll end up making personalized recommendations for purchases that you have
39:06
Ejaaz: within your conversations.
39:07
Ejaaz: And the reason why I think this approach is also really powerful is it's not
39:12
Ejaaz: kind of bothered about you coming onto Visa, setting up an account,
39:16
Ejaaz: attaching your agent to this thing.
39:18
Ejaaz: It just integrates directly into wherever you're using AI, whether that's Claude,
39:23
Ejaaz: whether that's ChatGPT, or whether that's even Meta.ai.
39:26
Ejaaz: And so two different approaches from two of the biggest companies on this play.
39:30
Ejaaz: You know, one thing it reminded me of, guys, is you know how when stablecoins
39:35
Ejaaz: became a pretty major thing, and it continually becomes a major thing with every
39:40
Ejaaz: week passing, Visa came in and said, Hmm, what can we do here?
39:45
Ejaaz: Okay, okay, okay. I get that you guys like stable coins. And I get that it's
39:49
Ejaaz: basically replacing the dollar and it's quicker and it's better than Swift. We'll support that.
39:54
Ejaaz: Just let us take a tiny fraction percentage of this transaction flow and we'll
39:59
Ejaaz: be good. Does that sound good?
40:01
Ejaaz: And everyone said, okay, yeah, that sounds good. We're saving money.
40:04
Ejaaz: And they ended up making like $500 million in the first year.
40:08
Ejaaz: Now imagine applying that to just anyone performing any kind of economic activity?
40:14
Ejaaz: You know, what percentage of David's Amazon grocery list do I want per year?
40:19
Ejaaz: You know, how much is that worth for me times, you know, all his activities that he does?
40:24
David: I do want to defend crypto for a moment because we've just talked about how,
40:27
David: you know, with both Visa and MasterCard, it's very obvious, like the incentives
40:32
David: point towards Visa and MasterCard entering the AI space.
40:36
David: All AI commerce Visa wants, all ai commerce mastercard wants that's their business
40:40
David: model they get 2.9 or 3.5 on every transaction so they need to be able to give
40:45
David: ai's credit cards so that they can take that fee that's their business model
40:48
David: makes total sense this is totally expected um
40:51
David: and so like the yeah the notion of just like oh yeah crypto is for ai agents
40:55
David: uh because like both are code-based uh
40:58
David: i think still makes sense mainly because there are compliance things
41:02
David: that visa and mastercard have to deal with that the crypto space does not have
41:06
David: to deal with uh they have to deal with chargebacks and fraud and the bank secrecy
41:10
David: act crypto does not have to do that uh
41:14
David: irrevocable transactions means that we don't have to worry about chargebacks
41:18
David: and fraud and all of those things you just have to be far more careful about
41:22
David: the transactions that you make which is up to the ai developers in the crypto
41:26
David: space and so there are fundamental
41:28
David: breaks on the trad AI commerce space that crypto will not experience.
41:35
David: And, you know, the idea of there being 10 billion AI agents all making commerce,
41:41
David: that doesn't mesh well with fraud chargebacks and the Bank Secrecy Act.
41:46
David: And so if we're truly trying to have scale and the number of agents who are
41:51
David: able to do commerce freely without frictions, and, you know,
41:54
David: 2.9% is also a lot, especially if we go into high volume
41:58
David: microtransactions, there is actually plenty of room for the efficiencies of blockchain rails.
42:03
David: And it's not to say that like, oh yeah, AI agents can use blockchain rails to
42:07
David: like commit fraud and get around the bank secrecy act.
42:10
David: It's just about the inversion of the rules where because there are chargebacks
42:15
David: on Visa, there therefore needs to be fraud departments and like all of these things.
42:21
David: So it's not about like AI agents can commit fraud on blockchains uh they can
42:26
David: they probably will i should probably assume should assume that but it's still
42:29
David: going to be like a minority of global use cases just because of the nature of the
42:34
David: different properties between blockchain commerce and uh visa mastercard commerce
42:39
David: and so there i still think there is plenty of room for
42:42
David: AI blockchain commerce to exist. All right. So this week, I think is the first
42:46
David: week that we are talking about robots, but I think there's a natural synergy
42:51
David: between artificial intelligence and robots.
42:53
David: I don't think I need to explain how those two things go together,
42:57
David: other than it's scary and I'm scared about it. But Josh, maybe tell me what's
43:00
David: going on in the world of AI-ridden robots.
43:02
Josh: I actually, because this is the first time we're talking about robots,
43:05
Josh: I kind of want to set the stage of how big of a deal robots actually are.
43:09
Josh: I think a lot of people see them Yeah, I'm like, oh, this is cute.
43:12
Josh: Like I have a Roomba with arms now and like maybe you can get my groceries.
43:15
Josh: But like the opportunity is way, way bigger than that. And to explain,
43:19
Josh: I want to use Apple as an example.
43:21
Josh: When Apple released the iPhone 2006, I think their market cap was about $70 billion.
43:27
Josh: And at the time, ExxonMobil was the biggest company in the world.
43:29
Josh: They were like $350 billion, not even a trillion dollars.
43:32
Josh: So the perceived upper bounds of Apple, which created these like toy consumer
43:37
Josh: devices that were not really that serious, they were just kind of always like
43:39
Josh: talk to each other, was about whatever the multiplication is on that.
43:44
Josh: Like, let's say a 5x to get to the most valuable company in the world,
43:46
Josh: which was oil and oil ran the world. So that was clearly the max.
43:49
Josh: That was clearly the limit.
43:50
Josh: But fast forward to today, Apple actually hit a $4 trillion market cap because
43:54
Josh: it created an entirely new industry on top of the economy that actually was
43:57
Josh: worth a lot more than oil.
43:59
Josh: The same thing is kind of happening with robotics here. I think you could kind
44:03
Josh: of view consumer or like productive output of an economy based on the productivity
44:10
Josh: per workforce times the workforce.
44:12
Josh: And what we're going to see now with robots is that the workforce number will increase.
44:17
Josh: Will expand exponentially. And now our workforce will be reflective of not just humans, but robots.
44:23
Josh: And the thing with robots is they are much cheaper than human beings.
44:26
Josh: So when you are going to your board and you are explaining to them why you want
44:29
Josh: to keep these humans, it's going to be a very difficult argument to have when
44:33
Josh: the cost per robot is significantly cheaper.
44:36
Josh: So there's this forcing function of like, hey, robots are much better,
44:39
Josh: they're much cheaper, and they create a lot more productive output than us.
44:42
Josh: And also they will decrease the cost of goods because the actual cost of employment is so much lower.
44:47
Josh: So robots is a really big deal. And I think we'll probably see like the perceived
44:50
Josh: limit now, let's say it's Apple at $4 trillion.
44:53
Josh: Some robotics company will exceed that. And it will exceed that aggressively
44:57
Josh: because it is replacing a human workforce times X. We don't know that multiple,
45:01
Josh: but we can produce robots much faster than we can make babies that are working.
45:04
Josh: And like that is a big multiplier that I don't think people are taking into account.
45:09
Ejaaz: Josh, steady your horses here for a second. Can I just say, if I also hadn't
45:14
Ejaaz: nerded out about this robot stuff over the last two weeks, I would think you're
45:18
Ejaaz: completely insane, dude.
45:20
Ejaaz: I think that the mental block that I... It sounds crazy, but you're so right.
45:24
Ejaaz: The mental block that I was trying to jump through myself when I was like literally
45:28
Ejaaz: reading these news updates and watching these cute and sometimes terrifying
45:32
Ejaaz: robots go berserk. I don't know if you guys saw that video, by the way,
45:35
Ejaaz: of the robot going berserk. I was like, these things aren't real.
45:39
Ejaaz: They're not agile enough. They're not like, surely this is all just CGI.
45:43
Ejaaz: But I think what you're trying to say is we're reaching a point where they're
45:46
Ejaaz: going to be able to basically replace a lot of what us humans can do.
45:50
Ejaaz: Not just that factory work, but like cleaning the dishes, going and running errands for us.
45:54
David: Even robots that are one third the speed of a human. If a human is three times
45:58
David: faster than a robot, that robot can still work 24 seven and produce the same
46:04
David: output as a human for no money if you own the robot.
46:08
Josh: This is happening. There is no reality in which the robots that you're seeing,
46:11
Josh: even if they are a little AI enhanced, will not exist.
46:14
Josh: They will absolutely exist. They will have human capabilities.
46:17
Josh: They will have narrow band robots. They will have general band robots with hands
46:20
Josh: and with the proper function that humans do.
46:22
Josh: This is happening. Absolutely.
46:25
Ejaaz: Josh, how much is this going to cost me? How much is it going to cost me? Is it the cost of a TV?
46:31
Ejaaz: Like a 52 inch TV back in the day?
46:33
David: What are we talking about?
46:35
Josh: Well, it depends on how it's going to cost you, but it comes in different ways.
46:38
Josh: So there's the cost of the actual robot. If you want a personal assistant,
46:41
Josh: that will hopefully start at like 40, 30 to $40,000. We're seeing a few early
46:45
Josh: versions of that and then rapidly decrease.
46:47
Josh: But I think a lot of the effects that people don't recognize is the robots that
46:50
Josh: are outside of your apartment, outside of your house, the ones that are working
46:53
Josh: in factories that are working significantly cheaper, but significantly more
46:56
Josh: efficiently that decrease the cost of goods.
46:59
Josh: So if we do have essentially an infinite workforce that is infinitely capable
47:04
Josh: and infinitely energized and infinitely patient,
47:06
Josh: Then they can be working 24-7, 100 times the efficiency that we are,
47:10
Josh: and decrease the cost of goods sold across every single medium.
47:13
Josh: So perhaps it costs you tens of thousands of dollars to get one in your house
47:17
Josh: initially, but that cost goes down.
47:19
Josh: And also the cost of goods that you buy outside of your house will go down significantly.
47:25
Josh: And this is kind of this paired with AI. So you have these smart robots that
47:28
Josh: are getting increasingly intelligent. They're able to replace the human need
47:32
Josh: throughout our economy.
47:33
Josh: We'll probably decrease the cost of a lot of things really, really rapidly.
47:38
David: And it's probably worth noting that there are already robots out there um waymo
47:43
David: is a robot that is a robot car and that is that is doing the job of a human
47:48
David: person driving other people around and then i've also seen
47:53
David: um i think we've all seen like these little robot like uber eats food carriers
47:56
David: where you put the you put the food in the little robot the robot drives just
48:00
David: wheels itself over to the the destination and then it drops off the food and
48:04
David: it's also got a little face on it but it looks like a
48:06
David: box with wheels nonetheless a robot and that's what drosh is talking about with
48:10
David: these differently form factors for robots and so there's already robots out there
48:13
David: and i think we are seeing the possibility of them approaching a more humanoid
48:17
David: form factor why a humanoid form factor well because the world is the world is
48:22
David: a human form factor because we've built the world to
48:24
David: to work with a humanoid form form factor and so that's to be expected i have
48:28
David: not i haven't seen seen personally a waymo but i know that they're all over
48:31
David: san francisco i haven't personally seen a food carrier robot
48:34
David: But I don't know if you guys have, but I'm willing to bet us three and all the
48:38
David: listeners over the next three years, we're all going to have our,
48:42
David: oh, there's my first robot that I see out in the wild moment.
48:45
David: And we're all going to pop our robot cherry by like, oh, a robot just delivered
48:50
David: me some product or service.
48:52
David: And now that's just going to become ubiquitous in a very short order after that.
48:56
Josh: A helpful framework for thinking about it is all of the AI that we talk about
48:59
Josh: that's really exciting on the show, robots are the physical manifestation of
49:02
Josh: that. So robots are AI applied to reality.
49:06
Josh: And I think the opportunity and the craziness that we're going to see of this
49:10
Josh: crazy AI that we discuss every single year applied to the physical meat space
49:14
Josh: that we live in is going to blow minds. And it's increasing as rapidly as AI is itself.
49:21
David: Hmm i feel like this is just turning into the doom show this is the
49:23
Ejaaz: Doom show yeah i don't think it's.
49:26
Josh: Really exciting to me that gets me fired up yeah we're gonna have a
49:28
Ejaaz: Bunch of robots hang on a second we're just talking about people you know meta
49:33
Ejaaz: disconnecting people and you know giving them ai friends and now what is it
49:37
Ejaaz: going to be an ai robot friend like hell.
49:39
David: Yeah i want a robot probably
49:41
Ejaaz: Yes yeah probably probably but again this is not the doomer show okay so,
49:46
Ejaaz: what talking about like first interactions with the robot i have seen that delivery
49:51
Ejaaz: career robot david and it looks super cute right um nvidia also like announced
49:56
Ejaaz: that they have this like you know cute little groot robot thing and it's like
50:00
Ejaaz: this tiny little thing that looks like that what's that
50:02
Ejaaz: what's that disney film guys that i'm i'm the one with like the cute robot that
50:06
Ejaaz: roves on mars or whatever it's called,
50:08
Ejaaz: wally yeah yeah okay i'm i'm mocking my age a little,
50:14
Ejaaz: bit Then there's the next level up, right?
50:18
Ejaaz: There's that Black Mirror episode from, I think, season one or season two,
50:21
Ejaaz: where it looks like a dog, but it's trapped with a machine gun on its back, right? I'm pretty sure.
50:26
David: I saw a video of this last week, a dog with a machine gun on its back.
50:28
Ejaaz: Well, that was going to be my next point. Like, China's already got,
50:31
Ejaaz: like, 50,000 of these things in, like, a training camp.
50:35
Ejaaz: And it's, like, practicing, like, shooting practice. I watched this entire video,
50:39
Ejaaz: didn't understand a single word, but I understood one thing,
50:42
Ejaaz: the concept of death, right? And I was like, Jesus, this is crazy, guys, right?
50:47
Ejaaz: But kind of like stepping back, to your point earlier, David, like,
50:52
Ejaaz: Yes, these robots are becoming more humanoid, right? But they're also becoming more intelligent.
50:58
Ejaaz: And to kind of bring this back to like the AI side of things,
51:01
Ejaaz: it's because people are pairing, you know, this AI, which is basically replicating
51:05
Ejaaz: human intelligence with a human kind of like one of a physical form that isn't
51:10
Ejaaz: just a human, it's a robot.
51:11
Ejaaz: One thing that I found interesting is it's not as simple as taking OpenAI's
51:17
Ejaaz: AI model and sticking it in a robot's brain.
51:20
Ejaaz: That actually doesn't work. You have to create different types of models which integrate,
51:27
Ejaaz: different multimodal mediums like vision and translating that into interpretability
51:33
Ejaaz: and then action and understanding and all that kind of stuff.
51:36
Ejaaz: It requires new models.
51:39
David: Right? OpenAI, ChatGPT, all these models, they're all thinking.
51:42
David: It's all thought. It's all cognitive. It's not about senses.
51:45
Ejaaz: Yeah, exactly. It's all characters. It's things that computers today understand, but robots have no idea.
51:51
Ejaaz: They don't know to look at a lamp next to you and be like, yeah,
51:55
Ejaaz: there should be a switch somewhere here that I can flick, right?
51:59
Ejaaz: So right now, I think we're at the GPT-1 or GPT-2 moment of models,
52:05
Ejaaz: right? So we've got NVIDIA releasing, I think the actual model is called Groot,
52:09
Ejaaz: which is like a general purpose thing, but it's very deterministic.
52:13
Ejaaz: So they're saying, hey, robot, when you see this water bottle,
52:17
Ejaaz: it's a water bottle and it's something that contains a liquid.
52:20
Ejaaz: So it's kind of like self-guiding.
52:22
Ejaaz: And then recently this week, I just saw a major update from this company in
52:27
Ejaaz: California, which basically released something called a 0.5 pi model,
52:32
Ejaaz: which is more of a generalist model.
52:34
Ejaaz: And they've put it into their own homemade robot and it can basically move around
52:38
Ejaaz: and it can understand kind of homemade tasks.
52:41
Ejaaz: It could see like your sink for the first time and see that there's dishes in
52:45
Ejaaz: there and be like, I should clean these dishes, like it's got done.
52:48
Ejaaz: So it understands and it interprets and it's kind of like accelerating at honestly
52:53
Ejaaz: a speed that I didn't think was possible like six months ago.
52:56
Ejaaz: Josh, do you have any further insight into how these robotic models work?
53:01
Josh: No, you're so right. A lot of models that we're used to today,
53:03
Josh: they use token-based text models.
53:06
Josh: And text doesn't really apply when you have eyeballs and you have ears and you
53:09
Josh: have sensors outside in the physical world.
53:11
Josh: There's this really great example that I love because it's so reflective of how early we are.
53:17
Josh: And it's the Tesla Optimus robot. And when they first trained it,
53:20
Josh: they didn't have any data on humans. They wouldn't strap cameras to humans' heads.
53:25
Josh: So what they did is they fed it the car's autopilot data. So for the first few
53:29
Josh: months of training, Optimus thought it was a car.
53:31
Josh: And it was viewing walkways as car lanes and it would look for stop signs.
53:36
Josh: And you kind of have to iterate through and train it like, hey,
53:39
Josh: you're not a car, but you live in the same type of reality.
53:42
Josh: And we're seeing exactly like you said, just very early versions of that where
53:46
Josh: the models are starting to get trained. They're starting to collect data,
53:49
Josh: but they haven't really experienced that takeoff moment that we've had in the.
53:53
Josh: General text-based models where every single week we're getting a new frontier model.
53:56
Josh: So I think the curve is slightly further behind, but I'm sure that's something
54:00
Josh: that we're going to be seeing a lot of is a race for data around real world's
54:05
Josh: inputs and multi-sensory, multimodal inputs.
54:08
Josh: And that's something that I think is super important to watch as these robotics
54:11
Josh: companies start to spin up humanoids like this one that we're looking at right now.
54:16
David: As we explore more and more subjects on Limitless, especially in the Monday
54:20
David: interview episodes that Josh and I are going to do, I think one of the big themes
54:23
David: is we are capable of using AI to accelerate innovation in other sectors.
54:29
David: And so generating robot models, we have the AI tools to do that faster now.
54:36
David: And so what might have taken a decade is going to take six months.
54:40
David: Everything's getting faster. And one of the other things that we frequently
54:42
David: bring up on this episode is, are any of your guys' friends using AI?
54:47
David: My crypto friends are, my non-crypto friends are not.
54:49
David: And so they are not even aware of the massive amount of intelligence that's coming into existence.
54:56
David: But they know about it, but robots are just not on their radar.
55:00
David: So telling them that like in three years, robots will walk among us and have
55:05
David: chat GBT level of intelligence, but they are still beginning to grok.
55:08
David: Parts of society are just completely blind to this.
55:13
Ejaaz: I saw this really interesting chart this week, guys. Actually,
55:16
Ejaaz: I think I saw it yesterday, which tracks the visit, visitation and usage of,
55:23
Ejaaz: OpenAI's ChatGPT website.
55:25
Ejaaz: And it's up only during Monday to Friday.
55:30
Ejaaz: And then it stagnates and goes down Saturday and Sunday.
55:33
Ejaaz: Now, there's many reasons why that might be the case. But the major takeaway
55:38
Ejaaz: that I've seen floating around is that people are using it for their work.
55:42
David: For work.
55:43
Ejaaz: Like 24-7 to do a bunch of different things. I speak to a lot of people that
55:46
Ejaaz: are in the non-crypto world that just use AI consistently to generate documents,
55:52
Ejaaz: PDFs or whatever it might be, like proposals or pitches for their kind of sales team.
55:57
Ejaaz: And I think we're just going to see this accelerate even more. like personally and.
56:01
Josh: Professionally a lot of my friends still use chat gpt as google it's an extension
56:05
Josh: of google and that's the extent of it i'd say without them maybe
56:09
Josh: a handful of people that i know actually use it in a further sense than that
56:13
Josh: um the professional thing is is only existent now because we don't have the
56:18
Josh: full stack to actually replace the job so they're kind of
56:21
Josh: they're being the leveraged humans but once that that gap is bridged where they're
56:26
Josh: no longer needed for to create these inputs it's probably the end
56:30
Ejaaz: Do you think people are going to keep critiquing it until suddenly it all falls in place,
56:36
Ejaaz: it feels like one of those suddenly and then all at once moments where people
56:40
Ejaaz: are just going to be like ah it's not good enough it's it's not going to replace
56:43
Ejaaz: me blah blah blah and then suddenly just does it immediately through like some
56:47
Ejaaz: kind of open ai model update and then,
56:49
Ejaaz: it's over basically and everyone kind of flips yeah.
56:53
Josh: Yeah, I think it's kind of what we're seeing with the chat GPT thing.
56:57
Josh: It feels like we have such a superpower being a part of the show or even listening
57:01
Josh: to the show where you're aware of what's happening.
57:04
Josh: So much of the world has no clue how quickly things are advancing.
57:07
Josh: And eventually there will become that killer consumer product,
57:09
Josh: the chat GPT moment for something that affects them.
57:13
Josh: And at that moment, they're going to be like, oh my God, where did this come
57:15
Josh: from? Where did this happen? Has this been going on? And the answer will be yes.
57:19
Josh: But for most people, they're just blissfully unaware of the rate of progress that's happening.
57:22
David: So while the rest of society continues to kind of be behind the curve of learning
57:28
David: about AI, I think for people who like us doing the show and listeners of the
57:32
David: show who are very aware of the growth of AI,
57:36
David: I think it's worth acknowledging that we are also behind the curve on one more
57:39
David: narrow aspect of AI, which is the idea of interpretability.
57:43
David: And so this subject has been going around downstream of this blog post from Dario Amodi.
57:49
David: This came out this month. I just pictured that name. I apologize.
57:52
David: But maybe walk us through this blog post, the idea of interpretability,
57:58
David: why it's important, what the problem is, and why we are behind the curve on it.
58:03
Ejaaz: Sure. So for those of you who don't know, Dario isn't some kind of peasant wandering around town.
58:09
Ejaaz: He is a co-founder of Anthropic, been in ML and AI research and product building
58:16
Ejaaz: for over a decade at this point, and one of the smartest people to be building within this space.
58:22
Ejaaz: And he released this blog post earlier this week around this concept called interpretability.
58:27
Ejaaz: And I'm going to say that word a lot, so I'm probably going to butcher it at
58:29
Ejaaz: some point. But it's this really interesting concept where I think to date,
58:34
Ejaaz: everyone thinks that AI does all this magical stuff, and you'd be right to think so.
58:40
Ejaaz: And so you might then think that the creators of these AI models would be able
58:44
Ejaaz: to explain how the model comes up with the answers that it gives, right?
58:50
Ejaaz: Would you expect that, David and Josh? Like you would expect like,
58:54
Ejaaz: okay, if a model is telling me something, I'm guessing if I go to Sam Altman,
58:57
Ejaaz: he'll be able to explain, yeah, it's because we tune this parameter.
59:00
Ejaaz: And that's why it's able to like give you this particular answer.
59:03
Josh: Intuitively, yes.
59:04
Ejaaz: Yeah, intuitively, right? But in reality, that is not the case at all.
59:09
Ejaaz: All they know is that when they put in these like training methods,
59:14
Ejaaz: and they input data into these models and they get an output,
59:18
Ejaaz: they don't actually know how the model thinks in between from the input to the
59:22
Ejaaz: output beyond the weights that they've kind of designed.
59:26
Ejaaz: And the analogy here would be, I think in software, you know what comes out
59:31
Ejaaz: of the system because humans deterministically
59:34
Ejaaz: code the paths that that software is going to execute on.
59:37
Ejaaz: But AI models are more like emergent organisms. They're kind of like a bacterial
59:42
Ejaaz: culture or when you breed racing horses, for example, you know,
59:46
Ejaaz: you can do your best to kind of combine the best traits for what the offspring is going to be.
59:50
Ejaaz: But at the end of the day, you have no control or idea what the product is going to be.
59:54
Ejaaz: And it's hard to predict the exact thoughts or perspectives that it's going to have.
59:58
Ejaaz: Now, the reason why there's very minimal research on this particular problem,
1:00:03
Ejaaz: on interpretability, is because it's hard to prove that there's a problem in the first place.
1:00:09
Ejaaz: So you can't show how the model thinks.
1:00:13
Ejaaz: So if you can't show how it thinks, how can you prove that it has nefarious
1:00:16
Ejaaz: or deceitful intent and this leaves us in a pretty dangerous precedence where
1:00:21
Ejaaz: we're like kind of like okay uh do we trust before verifying the models or only
1:00:26
Ejaaz: when they kind of fuck up do we then go ah.
1:00:29
Ejaaz: There's a problem and it's killed like, you know, half the human population,
1:00:32
Ejaaz: you know, maybe we should do something about it, right?
1:00:34
Ejaaz: Obviously, I'm exaggerating here, but this isn't something that is kind of unknown
1:00:40
Ejaaz: right now. So there's this familiar concept in AI models called chain of thought reasoning, right?
1:00:46
Ejaaz: Which is where the model gets a prompt and it like goes through its kind of
1:00:50
Ejaaz: reasoning process. I'm simplifying it quite a bit and probably Josh can get
1:00:54
Ejaaz: into the nitty-gritty of it.
1:00:56
Ejaaz: But what they recently found, and we spoke about this on last week's episode,
1:01:00
Ejaaz: is that the model was lying when it was all like, we went back and forth on
1:01:05
Ejaaz: like, it was wrong. It was wrong.
1:01:08
Ejaaz: And the reason why it was wrong was based off this concept of what it believed
1:01:11
Ejaaz: to be real, true, and reality.
1:01:14
Ejaaz: And the issue then comes, if you can't prove how an AI model thinks, and.
1:01:20
Ejaaz: You know, this model potentially is getting things wrong or lying,
1:01:24
Ejaaz: whichever way you want to look at it, then how can you ever detect what nefarious
1:01:29
Ejaaz: intent is for these models?
1:01:31
Ejaaz: And so we get into this kind of weird thing where it's like this AI is taking
1:01:34
Ejaaz: over more and more responsibility.
1:01:36
Ejaaz: We talked earlier about how these AI models are going to probably start to influence
1:01:40
Ejaaz: at the government and nation state level.
1:01:43
Ejaaz: Maybe we should have something in place to actually understand how these AI
1:01:47
Ejaaz: models kind of work, right?
1:01:49
Ejaaz: And so Dario and the Anthropic team have been focused on trying to create kind
1:01:53
Ejaaz: of like an MRI scan for AI models, but it's very much in its early days.
1:01:58
Ejaaz: And models have mechanisms very similar to like neurons in your brain,
1:02:01
Ejaaz: right? When they recognize a car or a horse, very specific neurons,
1:02:05
Ejaaz: they kind of like light up and say, hey, this is a car.
1:02:07
Ejaaz: This is a horse. This is a fire. It's hot. Don't touch it, right?
1:02:11
Ejaaz: And a group of the, I'm almost done with my my tutor session but a group of
1:02:15
Ejaaz: these neurons is known as like a feature.
1:02:17
Ejaaz: And emode detected 30 million features manually in a kind of medium-sized ai
1:02:24
Ejaaz: model but this was just manual
1:02:26
Ejaaz: but there's probably many many more right the ability to automate the process
1:02:30
Ejaaz: for detection would reveal all of these things and the thing with features so
1:02:34
Ejaaz: this group of neurons is that
1:02:36
Ejaaz: they give more insight into what goes on in a model's thinking so it starts
1:02:40
Ejaaz: to assess a prompt and all of these wonderful things but i'm just kind of thinking
1:02:44
Ejaaz: like we should care about this a lot more and i'm kind of curious why no one's
1:02:47
Ejaaz: kind of like raised this before josh do you have any any takes on this.
1:02:51
Josh: First that was that was brilliant i learned a lot you did such a great job of
1:02:55
Josh: describing this entire educate weird wacky world that's happening here
1:02:59
Josh: um to me interpretability feels like the the quantum physics to general physics
1:03:04
Josh: it doesn't really make sense it's really spooky and magical and no one really
1:03:08
Josh: knows what the hell is actually happening so that's
1:03:11
Josh: that's kind of where I'm at where I don't have any concrete answers or even
1:03:14
Josh: guesses at what the hell is going on.
1:03:16
Josh: It did remind me of this interesting thing about
1:03:19
Josh: transformers and large language models in general which is that they they are
1:03:23
Josh: at the end of the day they're token predictors
1:03:25
Josh: and the math that we can understand is the basic so basically when when a model creates another token
1:03:31
Josh: it does this matrix math through a transformer and that math outputs the next token
1:03:35
Josh: but before that there is these new models will have two trillion different parameters
1:03:40
Josh: that are all given different weights that result in that one single token so
1:03:43
Josh: to reverse engineer two trillion different parameters and to understand
1:03:47
Josh: the the matrix behind how they work That seems almost incomprehensible.
1:03:52
Josh: And I'm sure there might be interesting ways that there's like now you're talking
1:03:54
Josh: about neurons and features and and these things are all very foreign to me.
1:03:58
Josh: So I'm glad Dario is the one who is taking charge of this. He seems very bright.
1:04:04
Josh: He is the anthropic guy. He's probably well equipped to tackle this.
1:04:08
Josh: It's just weird. And it seems incredibly important because as these things get
1:04:13
Josh: more influential, as these things impact more of our lives, we want to understand how it works.
1:04:18
Josh: But like, I just have no idea how.
1:04:21
Ejaaz: Well, how do you how do you rebuild intelligence without even knowing the human brain in its entirety?
1:04:27
Josh: Right? Yeah.
1:04:28
David: I'm seeing a lot of parallels here to the understanding of the functioning of the human brain.
1:04:33
David: Specifically, we're talking about the domain of cognitive psychology,
1:04:35
David: right? Like if you take a cognitive psychology class, you'll just learn how
1:04:39
David: the brain operates as a computer, like cognitively.
1:04:43
David: Like some things are the motherboard, some things are the GPU.
1:04:46
David: Now here's like the eyeballs and how everything kind of like fits together.
1:04:50
David: And when you also take a like a when you learn about mental health psychology
1:04:55
David: you will learn about clustering of thought patterns or structuring of neurons
1:05:00
David: that relate to each other in ways that are atypical
1:05:03
David: that result in maladaptive outcomes from the person itself and so what i'm seeing
1:05:08
David: is dario is attempting to
1:05:10
David: identify clusters of parameters which is all all there's this ancient idea in
1:05:15
David: psychology called neurons that fire together wire together so
1:05:18
David: two neurons fire
1:05:21
David: uh and they they identify the firing of a an approximate local
1:05:25
David: uh adjacent neuron and when they are firing at the same time they start to move
1:05:29
David: closer to each other and that's how habits get formed that's how knowledge gets instantiated
1:05:36
David: And this is all how like good outcomes and bad outcomes for whatever they are,
1:05:39
David: like patterns get established.
1:05:41
David: And so if there is a lying or a deceitful or
1:05:45
David: like consistently incorrect chat GBT, there's going to be maybe a clustering
1:05:50
David: of parameters that represents a maladaptive outcome that it learned from its training.
1:05:57
David: And so I think there's a lot of parallels going on here.
1:06:00
David: And I think what we are really doing is we are trying to map the brain of an LLM.
1:06:05
David: However, I don't know how this process works. But we once upon a time mapped
1:06:10
David: the brain in terms of neurons.
1:06:12
David: And now we have a map of the brain. And we know what parts of the brain deal
1:06:15
David: with sight and what parts of
1:06:16
David: the brain represents your foot and what parts represent fear and memory.
1:06:22
David: And we just did that as a manual iterative process and I think we're just going
1:06:25
David: to do the same thing with mapping the parameters of an LLM if that's even possible,
1:06:31
David: which I don't know why it wouldn't be.
1:06:33
Ejaaz: Yeah, that's effectively it. And that's this whole kind of MRI scan that Dario
1:06:39
Ejaaz: keeps referencing throughout the piece, that he's trying to basically assess
1:06:43
Ejaaz: the different parts of the model and what relates to what kind of output.
1:06:47
Ejaaz: One interesting thing that he mentions in the post is that he's betting that,
1:06:53
Ejaaz: this MRI scan of interpretability will be achieved within five to 10 years.
1:06:59
Ejaaz: And in my opinion well in my opinion it's because there's so little research
1:07:04
Ejaaz: and that is an issue if the 2027,
1:07:10
Ejaaz: agi article right two episodes ago comes out in two years so
1:07:14
Ejaaz: the point he makes is that damn this is a by the way this is a problem we should
1:07:17
Ejaaz: really focus on and we should really get it done before agi is achieved because after agi is achieved
1:07:25
Ejaaz: there's there's no you can't The door is open. It's done. There's no one doing it.
1:07:29
David: The window of plasticity has shut. I think we should all kind of take a step
1:07:34
David: back and reflect on what this AI industry is doing when you zoom all the way back out.
1:07:39
David: And there's an idea out there that what is the healthcare industry doing?
1:07:43
David: You have doctors that are trying to cure cancer. There are other doctors that
1:07:48
David: are trying to cure heart attacks. There are other doctors that are trying to
1:07:51
David: fix Lou Gehrig's disease.
1:07:53
David: When you sum it all together, what is healthcare doing? And the answer is,
1:07:57
David: it is trying to learn how to make us live forever. It is trying to fix all disease.
1:08:02
David: Not any one doctor thinks that they are trying to make anyone live forever,
1:08:06
David: but if you are sick and dying, you go to the hospital and they try and stop you from dying.
1:08:11
David: And when you sum it, the whole entire vertical of health, it is trying to figure
1:08:15
David: out how to win longevity.
1:08:17
David: I think we can apply that same structure of thought to AI.
1:08:21
David: What is all AI trying to do? what are all AI models and AI labs and all of this stuff trying to do.
1:08:26
David: We are trying to create life. We are trying to create a new form of life,
1:08:30
David: a secondary non-carbon-based life form.
1:08:34
David: And all of these things are coming together. And so this idea of the cognitive
1:08:38
David: psychology of AI models, I think it's going to become extremely important because
1:08:42
David: that is our ability to understand this life form that we are creating.
1:08:47
David: And we, through this black box model of creating AI, are leaving bugs in the parameters.
1:08:54
David: There are exploits and bugs and lies and moral imperfections left in these parameter sets.
1:09:00
David: And we need to like go in and like fix those things before that window of plasticity
1:09:04
David: shuts and the way that life exists is the way that it will exist and it's like
1:09:08
David: we are unable to undo it. That's what I see when I see this.
1:09:13
Josh: That's good. We're kind of programming the DNA of the next form of intelligence.
1:09:17
Ejaaz: Well, let me ask you guys this. Whose lives matter more?
1:09:24
David: I don't think you can moralize about it. Life is life.
1:09:27
Ejaaz: Yeah. Yeah.
1:09:29
David: I mean, AI life doesn't exist yet. So right now, humans, and we should be aware
1:09:33
David: of that. But in the future, there is going to be indistinguishability between
1:09:37
David: what life means between whether it's carbon or silicon.
1:09:41
Ejaaz: And if you have this hyper-optimized form of life that will almost always outcompete the,
1:09:48
Ejaaz: pudgy, flesh-ridden humans, then maybe we just end up living in a world full
1:09:54
Ejaaz: of hybrid human robots or just robots on its own.
1:09:57
David: The hybrid human relationship i think is the best outcome and the outcome worth
1:10:01
David: striving for the because there's also the outcome where uh it's just the robots
1:10:05
David: and the humans are are ants as we are ants as it relates to ants to humans except
1:10:11
David: the robots are the humans now
1:10:12
Josh: The the human robot connection feels like base case um that feels like the that
1:10:17
Josh: will certainly happen And we're seeing that with neural link brain to machine
1:10:20
Josh: interfaces that is happening.
1:10:22
Josh: It's the, it's the, are we a bootloader for intelligence? And we appreciate
1:10:26
Josh: our meaning that is like the scary case.
1:10:29
Josh: So it will exist somewhere along that spectrum, but base cases,
1:10:31
Josh: absolutely. We, we merge with this stuff because it will be so superior to us.
1:10:36
David: All right, guys, we are going over on time. So, Jaws, I want you to run us through
1:10:40
David: what we are calling the dopamine section.
1:10:42
David: So I'm going to read out what are the words here. AI, ASMR, AI agents are redacted.
1:10:49
David: The girlies are asking GPT for a glow of advice and IQ of AI has jumped 40 points
1:10:54
David: in one year. Speed run us through all these subjects.
1:10:56
Ejaaz: Okay, let's hit the first one. So AI ASMR.
1:11:01
Ejaaz: The point that this video is demonstrating is that both video and sound AI models
1:11:07
Ejaaz: are getting really, really good.
1:11:09
Ejaaz: But rather the product of this video is, for those of you who can't see,
1:11:14
Ejaaz: is a gingerhead Caucasian woman sitting in front of a podcast mic,
1:11:19
Ejaaz: and she's speaking clearly into it, but she sounds really, really human.
1:11:23
Ejaaz: And for those of you who have watched ASMR videos, you'll get the idea of what
1:11:26
Ejaaz: that sounds like in your headphones.
1:11:28
Ejaaz: And she's basically trolling a bunch of AI researchers.
1:11:32
Ejaaz: So it's real like nerd fest right now, but she talks about, you know,
1:11:37
Ejaaz: fine tuning a different data set and all these like nerdy things.
1:11:40
Ejaaz: But it's just incredibly realistic how like these things are becoming.
1:11:44
Ejaaz: And I thought that was pretty funny to watch.
1:11:47
Ejaaz: The next one that's coming up is AI agents are redacted.
1:11:52
Ejaaz: Now, this is a Carnegie Mellon University company that was started only with AI agents as employees.
1:12:01
Ejaaz: Now, if you kind of think it went really intelligently and well,
1:12:04
Ejaaz: because, you know, AI models are incredibly intelligent, you would be wrong.
1:12:10
Ejaaz: The best performing employee, which was an AI agent, which was Claude specifically,
1:12:15
Ejaaz: only completed 24% of tasks that was set forth for it.
1:12:20
Ejaaz: And the tasks that it was given were things that a normal employee working at,
1:12:25
Ejaaz: your average sized company would do.
1:12:26
Ejaaz: So reading emails, maybe doing some coding, taking some calls,
1:12:31
Ejaaz: messaging other employees to say, hey, here's the update from my end.
1:12:34
Ejaaz: And this simulation ran and basically the takeaway is that we're not,
1:12:39
Ejaaz: quite there just yet. But, you know, it's a it's a funny observation.
1:12:43
Ejaaz: And I want to check back in in about six months time when these agents are probably
1:12:47
Ejaaz: way, way, way more intelligent.
1:12:50
Ejaaz: The third thing here is, I saw this post the other day, and I kind of like laughed
1:12:54
Ejaaz: because I think like, my kind of interactions with AI has been kind of similar,
1:12:59
Ejaaz: but just from a different perspective.
1:13:01
Ejaaz: But it's titled the girls are using chat GPT for glow up recommendations.
1:13:06
Ejaaz: And the results are pretty good. So what we see here in the snapshot is this
1:13:11
Ejaaz: girl asked ChatGPT, how can I improve my appearance? And she just posts a picture of herself.
1:13:16
Ejaaz: And then ChatGPT gives her an AI edited glow up version of what she could look like
1:13:23
Ejaaz: with annotations of what she could do to herself, like dye your hair,
1:13:27
Ejaaz: chocolate brown, use a peachy lipstick and blush and use bronze eyeshadow.
1:13:33
Ejaaz: And then she did all those things that it suggested and posted her glow up feature there.
1:13:38
Ejaaz: And it got apparently like a pretty crazy response from people being like,
1:13:41
Ejaaz: hey, this is like pretty cool or whatever. So I don't know about you,
1:13:44
Ejaaz: but I'll probably stop looking in the mirror and just start doing this going forwards.
1:13:50
Ejaaz: And the final point here is the IQ of AI has jumped 40 points in one year.
1:13:56
Ejaaz: Now, this is basically a measure of the IQ of these different AI models.
1:14:02
Ejaaz: And if you were to extrapolate this out going forwards, it's basically these
1:14:06
Ejaaz: things are going to become much smarter than humans on average.
1:14:09
Ejaaz: And this is an on average take in probably about a year and a half time, right?
1:14:14
Ejaaz: And whilst this isn't like kind of like a fancy, cool thing to look at,
1:14:18
Ejaaz: it's just something to keep in mind that these models are getting way more intelligent than you think.
1:14:22
Ejaaz: And for all the critics that are saying, hey, it doesn't understand the nuance
1:14:26
Ejaaz: of this, or it just doesn't understand my personality, we're going to reach
1:14:29
Ejaaz: a point where these AI models and AI agents understand you way better than you
1:14:32
Ejaaz: understand yourselves.
1:14:33
Ejaaz: And that should be expected more imminently than it is a far off kind of dream.
1:14:39
David: Maybe to drive the point at home about how big 40 IQ points are,
1:14:43
David: 10 IQ points is one standard deviation.
1:14:46
David: Four standard deviations, because of 40 IQ points, means that AI models have
1:14:51
David: surpassed from going from the bottom 0.003% of the population to the top 99.997% of the population.
1:14:58
David: That happened this year.
1:15:01
Josh: In one year.
1:15:02
David: In one year.
1:15:03
Ejaaz: Crazy. Wow.
1:15:05
David: Guys, we covered a ton in this AI rollup. I love these episodes.
1:15:09
David: I learned a lot from you guys. See, Joss, thank you for helping us put the agenda
1:15:11
David: together. And Josh, thank you for your takes as always.
1:15:15
Ejaaz: Awesome.
1:15:15
Josh: It was a pleasure. Another great week.
1:15:17
David: Yeah, this is no longer the Bankless Podcast, so I don't know if I have to give
1:15:20
David: a crypto disclaimer. This is the Limitless Podcast. The future is weird.
1:15:23
David: The future is risky. And that is why we are doing these episodes to help us
1:15:26
David: all stay ahead of the curve.
1:15:28
David: And we are glad you are joining us on this journey into the frontier of technology
1:15:31
David: and AI. So come back next week.
1:15:33
David: Subscribe to the podcast if you have not already. Subscribe to the YouTube if
1:15:36
David: you have not already. make sure to give us a five-star review so we can grow
1:15:39
David: this podcast and push it to the frontier of podcasts where this podcast deserves
1:15:43
David: to be limitless nation i guess uh we'll see you in seven days
1:15:48
Ejaaz: See you guys see you.