Latest / In the Loop AI / Understanding The Past Year in AI: 200th Episode Special!

Transcript

Understanding The Past Year in AI: 200th Episode Special! · In the Loop AI

0:00
Ejaaz: Welcome officially to our 200th episode.
0:04
Ejaaz: 1.4 million words has been spoken by us and 99.9% of it has been Josh and I
0:09
Ejaaz: arguing about who has the most compute, who has the better AI model,
0:13
Ejaaz: or which is the best AI company to invest in.
0:15
Ejaaz: Now, before we started this episode, we did something a little dangerous.
0:18
Ejaaz: We gave Claude and ChatGPT access to our entire 200 transcriptions for all of our episodes.
0:25
Ejaaz: And we got it to identify any hidden investment themes in AI that we might potentially
0:30
Ejaaz: have missed, all our favorite quotes and bets that we've made over the last
0:34
Ejaaz: year, as well as where the AI puck is going.
0:36
Ejaaz: This is going to be a jam-packed episode of information that we ourselves didn't
0:40
Ejaaz: know, but Claude and ChatGPT helped us figure out.
0:43
Josh: It's pretty cool. We gave our entire episode database to this model and said
0:48
Josh: like, hey, kind of roast us, come up with some themes, come up with some trends.
0:51
Josh: And just to backtrack a little bit of the world 14 months ago,
0:55
Josh: when we first hit record on the first episode and how much has changed.
0:58
Josh: I mean, at that time, SpaceX was private. They were worth $350 billion.
1:02
Josh: Now they're public worth over $2 trillion. Anthropic was worth $61 billion.
1:08
Josh: Now they're worth roughly a trillion. I mean, the multiples on these companies
1:11
Josh: are huge. This is when OpenAI was worth only $300 billion. ChatGPT had about half a billion users.
1:16
Josh: That has since tripled and doubled. And the models, the funniest thing is the models.
1:21
Josh: GPT4R0 is the flagship model. We had Cloud 3.7 and Gemini 2.5.
1:25
Josh: And then most surprisingly, everything was coded by hand.
1:27
Josh: At the time of first recording the first episode, nearly all code written around
1:31
Josh: us that created all the software in the world was written by hand.
1:34
Josh: The inverse is true. So so much stuff has happened over the last 14 months.
1:38
Josh: It's kind of hard to wrap your head around, But that's what we're going to do
1:40
Josh: now, is we're going to cover retroactively the things that have happened over
1:44
Josh: the last year, and then proactively where things are going based on all the
1:47
Josh: trends that we've evaluated over the last 14 months.
1:50
Ejaaz: So firstly, I want us to take a victory lap on the bets or predictions that we made
1:56
Ejaaz: early on in our career, back all the way back in 2025, that ended up becoming
2:00
Ejaaz: true both from an investment perspective and as a general trend playing out in the AI market.
2:06
Ejaaz: And then what we're going to get into is what that investment portfolio would
2:08
Ejaaz: have looked like if Limitless had hundreds of millions of dollars and we could
2:12
Ejaaz: have invested in some of these companies to start off with.
2:14
Ejaaz: So Josh, if you remember week number four, June 5th of 2025,
2:20
Ejaaz: we said verbatim, compute is now a matter of national security.
2:23
Ejaaz: Whoever owns the watts, the chips,
2:26
Ejaaz: the energy, the cooling owns the economic rents. The point that we were making
2:29
Ejaaz: back then was essentially compute is king.
2:33
Ejaaz: Whoever owns the GPUs, whoever can power them will build the best model.
2:37
Ejaaz: Now, fast forward 200 episodes later, this is more true than ever.
2:41
Ejaaz: Anthropic has signed, I think, four new compute deals over the last couple of months.
2:45
Ejaaz: You've got OpenAI expanding aggressively. Their big bet on acquiring as much
2:49
Ejaaz: compute as they can has turned out to be really, really good.
2:51
Ejaaz: They're not putting any kind of restrictions or limits on any users for any of their models.
2:55
Ejaaz: And it has become the key thing that is serving this thing called inference,
2:59
Ejaaz: which is helping people become smarter, run their AI agents 24-7.
3:03
Ejaaz: Computer's king. That was one of our first bets that we made.
3:06
Josh: You know what I just looked up is at the time that we said that, SanDisk is up 3,500%.
3:12
Ejaaz: Let's go.
3:13
Josh: Some crazy outrageous number. But yeah, it was so true. It was really early and true.
3:17
Josh: And I think that's still the case. And if you told the people from June 2025 that
3:21
Josh: the federal government was going to be invested in these companies like Intel,
3:24
Josh: that the amount of run-ups were going to be this large, like 35 times your money,
3:28
Josh: well, certainly we would have invested, but we also would have been pretty surprised
3:32
Josh: at how important this race to compute is.
3:34
Josh: The most scarce resource in the world right now is this energy to power on these
3:38
Josh: GPUs and to get memory for them. And to notice that early was amazing. I wish we acted on that.
3:44
Ejaaz: I know. The other asymmetric bet that we made was on AI chips, specifically memory.
3:51
Ejaaz: Now, Josh, if you remember, we had a predictions episode at the end of last
3:55
Ejaaz: year, and we had this kind of general thesis, which is memory is quite expensive.
4:00
Ejaaz: It's a large component of the whole entire AI thing. You need to power your
4:03
Ejaaz: GPUs, you needed to remember everything that you have in your conversation with
4:06
Ejaaz: ChatGPT and Claude, this will probably end up going up higher in terms of pricing,
4:11
Ejaaz: which means that the memory stocks themselves will probably perform very well.
4:15
Ejaaz: Since then, guess what the average
4:16
Ejaaz: return has been on the top three memory manufacturers? Just have a guess.
4:20
Josh: A gazillion dollars? It's infinite. It must be the single best investment that
4:25
Josh: you could have made in the last year.
4:26
Ejaaz: It's not a gazillion, but it's pretty close. It was 153%. If you'd bought Micro
4:31
Ejaaz: by the end of last year, you'd be up around 180%.
4:34
Ejaaz: If you had bought SK Hynix, if you were in Korea, if you had access to Samsung
4:37
Ejaaz: as well, SK Hynix has become the most valuable company in Korea.
4:40
Ejaaz: It's eclipsed Samsung, who's been kind of like the monopoly for the entire last
4:44
Ejaaz: couple of decades. It's just very, very impressive.
4:46
Josh: Actually, Samsung just became the most profitable company in the world over NVIDIA.
4:50
Ejaaz: Oh, yes. I saw that. I saw that.
4:52
Josh: So it's crazy. There's this all-out war and this race to the top,
4:56
Josh: and these companies are just like continuing to topple each other over and over and over again.
5:00
Ejaaz: Okay, so if it's not investments, Josh, I want us to kind of like have like
5:04
Ejaaz: a thesis call that we kind of nailed.
5:07
Ejaaz: Now, in March 20, 2026, so not too long ago, we said, this might be a really
5:13
Ejaaz: asymmetric bet, but I wish I could buy OpenAI stock right now,
5:16
Ejaaz: because there's a company that could sit on top of that.
5:18
Ejaaz: And it's called Cursor. What's happened since then to.
5:22
Josh: That little company? That's so nuts. three months later that same company got
5:26
Josh: purchased by spacex for 60 billion dollars my god should we just like we need
5:32
Josh: to start running a funder we need a
5:34
Ejaaz: Fund does anyone want to like.
5:35
Josh: Invest with us if we just invested the money that we had into these stocks we probably would have had
5:40
Josh: a pretty sizable portfolio to begin with but there is a really exciting segment
5:44
Josh: that i want to get into now that i have been waiting for so long to talk about
5:47
Josh: which is the limitless portfolio if those of you for those of you who've been
5:50
Josh: around for a very long time you know that in the early days of Limitless,
5:53
Josh: we actually recorded a lot of interviews.
5:55
Josh: We had on guests on the show who were building really cool frontier technology.
5:59
Josh: And those were very particularly selected. We were excited about a few specific companies.
6:04
Josh: Those companies we reached out to got the CEOs, the founders on the show.
6:07
Josh: And if we took all those companies and we put money in them on the day that
6:10
Josh: we recorded with those founders to today, the total return on that investment
6:14
Josh: would have been about 4x in just over a year, which I think is really noteworthy. It would have taken
6:20
Josh: A 13x multiple on our best performer, which is Valor Atomics.
6:23
Josh: If we go back to episode 10, we had the CEO of Valor Atomics,
6:27
Josh: Isaiah Taylor. For those who haven't watched, we actually reposted it because it was so good.
6:31
Josh: And I would recommend checking it out because he was the person who's responsible
6:34
Josh: for building small modular nuclear reactors.
6:36
Josh: The obvious flaw here was that data centers don't have enough energy and his plan was to solve it.
6:41
Josh: We had them on just after their seed round, and now they're valued at $2 billion, a 13x multiple.
6:46
Josh: Same thing happened with OpenRouter and the founder, Alex Atala.
6:49
Josh: At the time, they were half a billion dollars now they are 1.3 same thing with
6:53
Josh: zipline drone delivery that was a really cool one where they were showing early
6:58
Josh: prototypes of what the drone delivery
6:59
Josh: system looks like and now it's actually being used in a lot of major cities
7:03
Josh: same with perplexity 1.2 and then boom supersonic is the one anomaly they're
7:07
Josh: the company that was making supersonic airplanes if you remember
7:10
Josh: they've actually pivoted to making turbines so they're making supersonic airplanes
7:14
Josh: but now they're also making gas turbines for these ai data centers so i thought
7:17
Josh: this is really cool to see how large of a return we've had on some of the guests
7:21
Josh: that we've had on the show.
7:22
Josh: And if we just invested the Limitless Fund on day one of having all these people
7:26
Josh: on, we would have returned about 4x in the last year.
7:29
Josh: And that's only with private market valuations. Unfortunately,
7:32
Josh: none of these are public, but all pretty interesting investment opportunities.
7:36
Ejaaz: Yeah, I think it's amazing to think about like a 400% return relative to like
7:40
Ejaaz: the average market, which has gone up pretty insanely, but that would have been,
7:43
Ejaaz: Lovely to get access to. I would have taken that. I would have taken it.
7:46
Ejaaz: I think another thing that this theoretical portfolio shows us is the key trends
7:53
Ejaaz: that have developed over the last year.
7:55
Ejaaz: And I want to spend a moment to describe them very quickly, because I think
7:59
Ejaaz: it shows us where the puck is going in AI and from an investment perspective
8:03
Ejaaz: and from just a general trend and model usage perspective going forward.
8:07
Ejaaz: So with Valor Atomics, it's an energy thing, right? So you have all these GPUs,
8:12
Ejaaz: you have all these expensive data centers, but we can't quite get the power
8:15
Ejaaz: to get access to a bunch of these different things.
8:19
Ejaaz: Nuclear is an incredibly important facet of energy that will kind of like feed
8:23
Ejaaz: into a bunch of these things and eventually bring or run and operate a bunch of these things online.
8:27
Ejaaz: Now, on Open Router specifically, we've covered this thesis a few times, which is...
8:32
Ejaaz: We have all these different models. I don't think a company is going to get
8:35
Ejaaz: married to one particular model. I think they're going to use multiple models
8:39
Ejaaz: for multiple different things, use cases, etc.
8:41
Ejaaz: That's the whole point around the cursor thesis. That's why SpaceX acquired them for $60 billion.
8:46
Ejaaz: Open Router is pretty much the next best bet. They launch or stealth launch
8:51
Ejaaz: models before they actually become official. They've done it with Clawed.
8:54
Ejaaz: They've done it with ChatGPT. And they've done it with a bunch of Chinese models.
8:57
Ejaaz: And developers can get unrelentless access to all these types of models and
9:01
Ejaaz: use it for whatever that they want.
9:02
Ejaaz: Now, a key bit of data that they aggregate is intense, and they can use this
9:06
Ejaaz: intense to figure out what better models to build.
9:08
Ejaaz: Perplexity has absolutely nailed it. Like, look at this. I know it isn't the
9:11
Ejaaz: biggest multiple jump, but like $18 billion to $21 billion, this is primarily
9:16
Ejaaz: because Samsung has made their product the primary AI agent on all of their phones going forwards.
9:22
Ejaaz: It's just, you know, it's not just a portfolio return that we're talking about here.
9:26
Ejaaz: There's some major trends that have been developing that are feeding into what
9:30
Ejaaz: we might expect for an investment perspective going forwards.
9:33
Josh: So over the last 14 months, the trends have changed pretty considerably as well.
9:37
Josh: Things that we mentioned a lot now mostly no longer exist. And these turning
9:41
Josh: tides are interesting to see visualized here.
9:44
Josh: One of the things that we mentioned a lot in 2025 was superintelligence.
9:48
Josh: That's now down, what is that, 90%. We mentioned it just about 60 times last year.
9:53
Josh: We've mentioned it six this year. And this is about accurate because we recorded
9:57
Josh: for about six months last year, about six months this year. So it's right down the middle.
10:01
Josh: The biggest loser also is crypto, which we've mentioned basically zero because,
10:05
Josh: you know, it's in a bit of a winter right now, not doing super hot.
10:08
Josh: Robotics, surprisingly, has gone down 60% year over year, as well as AGI going
10:13
Josh: down 54%. So I think a lot of those broad sweeping terms, we mentioned less
10:17
Josh: probably because it feels like it's just close to being here.
10:20
Josh: And the lines have gotten so fuzzy that we're not actually sure what to even
10:23
Josh: define as AGI, what to define as super intelligence.
10:25
Josh: Like you look at these new mythos class models, you're like,
10:28
Josh: I don't know, that kind of looks like AGI to me, but the robotics part has been
10:32
Josh: interesting to me at least because it seems like we are in this like kind of...
10:36
Josh: Weird middle ground of robotics where a lot of companies are building them but
10:39
Josh: they haven't quite launched
10:40
Josh: what they're building just yet we have the new optimist robot that's coming
10:43
Josh: it hasn't been revealed figure has been working on their new robot hasn't quite
10:46
Josh: been revealed everyone's working behind the scenes and i assume those big reveals
10:50
Josh: are coming sometime later this year
10:51
Josh: and when that happens we'll probably have this big resurgence of robotics but
10:54
Josh: i found that really interesting another really interesting one was the word
10:58
Josh: anthropic oh my god what a year that company has had uh last year we mentioned them
11:03
Josh: i think one-fourth the amount that we did of open AI and chat GPT this year it's gone up 4x
11:08
Josh: we mentioned that company four times more and it's probably because I mean a
11:12
Josh: year ago things weren't looking super hot I remember we were looking at this
11:15
Josh: was like early opus like 3.7
11:18
Josh: Days and we were like yeah it's not that great like chat GPT just rocks it has
11:22
Josh: a really awesome consumer application and my god did this company turn around this year
11:27
Josh: It has now flipped OpenAI in terms of mentions on the show with 806 to 758,
11:33
Josh: Which is cool. I love that we get this granular ability to see all the stuff
11:37
Josh: using, ironically, Claude.
11:39
Ejaaz: These things move so quickly. Remember, Claude code is only,
11:42
Ejaaz: I think it's just over a year old, which is like mind-blowing because at the
11:45
Ejaaz: end of the last year, barely anyone used Claude code and now everyone uses it.
11:49
Ejaaz: There's also a lot of the few key trends that I've noticed have risen in the last couple of months.
11:54
Ejaaz: We're speaking about IPOs quite a bit on this show. We're talking about memory
11:58
Ejaaz: quite a bit. And it's unsurprising at this point because we have probably the
12:02
Ejaaz: biggest year for net new IPOs. We had SpaceX, we're expecting Anthropic,
12:06
Ejaaz: and OpenAI potentially by the end of this year.
12:08
Ejaaz: And then, of course, the memory companies, as we mentioned earlier,
12:11
Ejaaz: have been absolutely killing it on 80% margins at this point.
12:14
Ejaaz: Samsung just had a blowout quarter as well.
12:17
Ejaaz: You mentioned earlier being more profitable than NVIDIA. So a lot of these general
12:20
Ejaaz: themes are replacing some of the more vague themes that we had earlier.
12:24
Ejaaz: So AGI superintelligence, we don't quite know what that means.
12:27
Ejaaz: When we talk about like robotics, I think if I had to bet going forwards,
12:32
Ejaaz: that's going to flip over the next couple of months, because we have a bunch
12:35
Ejaaz: of new US-based robotic startups that are starting to sell their hardware to
12:39
Ejaaz: a bunch of different people going forwards.
12:42
Ejaaz: And then the anthropic flip is kind of expected. I think you and I use Claude
12:45
Ejaaz: pretty religiously at this point. I haven't used GPT for a while,
12:48
Ejaaz: rumored to be releasing GPT 5.6 Sol to the wider public, potentially today.
12:53
Ejaaz: If you're listening you might already have access to it, could be pretty cool.
12:57
Ejaaz: I think one of the most interesting charts is this one that I'm showing on the
13:00
Ejaaz: screen here, which is the company, the AI company specifically,
13:04
Ejaaz: the model lab that we have mentioned the most over the last 200 episodes.
13:09
Ejaaz: And it's surprisingly not OpenAI.
13:11
Ejaaz: It's not Anthropic, but it's Google. Do you remember Google?
13:16
Josh: Google is that company. Yeah, I remember obsession with Google.
13:18
Ejaaz: Yeah, yeah. It's this company that's kind of been around for a while.
13:22
Ejaaz: They dominate the search engine space.
13:24
Ejaaz: And they kind of are the only company that has a verticalized stack in AI.
13:28
Ejaaz: They create their own version of GPUs called TPUs.
13:32
Ejaaz: They have applications to disperse their AI models and they have their most
13:35
Ejaaz: famous AI model called Gemini.
13:37
Ejaaz: But Google's kind of fallen off since then. And I wonder if we see the flippening
13:41
Ejaaz: soon with either OpenAI or Anthropic.
13:43
Josh: I mean, I think those numbers are a little tweaked because we were obsessed
13:46
Josh: with Google for a few months and rightfully so.
13:47
Josh: They had some unbelievable product velocity. They were releasing things seemingly
13:51
Josh: every single week and they're all pretty hit features
13:53
Josh: that has since changed and yeah like you mentioned I think we certainly have
13:56
Josh: this flip happening where really the two main players in the space right now
14:00
Josh: are open AI and Anthropic and
14:02
Josh: that seems like it's gonna be that way for the foreseeable future perhaps we get
14:06
Josh: a comeback player of the year with Grok we'll see I know that the SpaceX AI
14:10
Josh: team which is now fully rebranded is working very hard to make this happen in
14:14
Josh: terms of this momentum map it's pretty interesting to see that we kind of have
14:19
Josh: These new emerging sectors that we haven't quite discussed just yet,
14:22
Josh: one of them being quantum.
14:24
Josh: Quantum seems to be the next lane, according to things that we've said,
14:28
Josh: and according to this chat of being the new thing that we're maximally excited
14:33
Josh: about. Things like drug discovery, biology, memory, IPOs, bubbles.
14:36
Josh: Those are all things that are very much accelerating.
14:39
Josh: The contrarian corner, which I'm not sure I agree with this totally,
14:41
Josh: has energy inside of it. I think energy is certainly accelerating as well.
14:45
Josh: Energy is, as always, the single bottleneck.
14:47
Josh: And even in the absence of ai acceleration energy is still the constraint to
14:52
Josh: pretty much everything that we do but noteworthy in the cooling sector we have
14:56
Josh: um open ai there which i was like a little disappointed to see i'm like man
15:00
Josh: what do you mean opening is cooling but then i think about the last couple months
15:03
Josh: it's been a little bit slow in open ai land they have these three new models
15:06
Josh: which they've been waiting to release i know the company is taking their summer
15:09
Josh: break this week they haven't quite gotten it out the government's been putting holds on it
15:13
Josh: the application hasn't quite gotten much better.
15:16
Josh: I know kind of after Codex for me, OpenAI has kind of fallen flat in terms of
15:20
Josh: their new feature releases. So you have to assume they're cooking up a lot of
15:23
Josh: stuff, but within the last month or two,
15:25
Josh: Which seems like the last year or two, they have been moving a little bit slower
15:29
Josh: than I think we would have liked. And I hope that they'll be able to turn that around pretty soon.
15:33
Ejaaz: Yeah, if I had to identify three themes that have become more apparent over
15:38
Ejaaz: the last couple of months that I think are important to keep an eye on going forwards.
15:42
Ejaaz: One is the increasing use of outer space to train models. Now,
15:48
Ejaaz: I remember the first time we mentioned StarCloud, which is this small startup
15:51
Ejaaz: that came out of Y Combinator.
15:53
Ejaaz: And they launched a bunch of GPUs into space and they started training models
15:58
Ejaaz: off of an H-100 set of GPUs from NVIDIA.
16:02
Ejaaz: Fast forward now, it's SpaceX AI's entire thesis that they're going to launch
16:08
Ejaaz: all these satellites out there and train models, presumably Grok and many others going forwards.
16:13
Ejaaz: This has become an increasingly strong theme that I think will play out more
16:16
Ejaaz: and more over the next couple of years. And I think SpaceX is going to be the one to lead it.
16:20
Ejaaz: The second thing is applications of these AI models into niche categories.
16:25
Ejaaz: Now, I think one thing that has become apparent to me is generalized LLMs are
16:29
Ejaaz: amazing. They're great chatbots.
16:31
Ejaaz: But when applied to specific niche sectors where you kind of need some expert
16:35
Ejaaz: knowledge, they aren't that very good.
16:36
Ejaaz: So when you look at Anthropic and OpenAI over the last couple of months,
16:39
Ejaaz: they've started their own joint ventures. They've raised billions of dollars
16:42
Ejaaz: to place their engineers in these companies and figure out the best solutions to build.
16:47
Ejaaz: I think the two sectors that they're going to be focused on going forwards is finance.
16:51
Ejaaz: And you mentioned earlier, the drug discovery, biology, science accelerated side of things.
16:55
Ejaaz: And I think we're going to see a bunch of really cool products and features
16:59
Ejaaz: being released over the next six months or until the rest of the year.
17:02
Ejaaz: You've got Anthropic that's already released, Claude Science.
17:04
Ejaaz: You've got OpenAI that's already released their gene benchmark.
17:08
Ejaaz: I think OpenAI is the one that has bought and operating retro science labs,
17:13
Ejaaz: which is working on a bunch of different cures. You've got Google that has as
17:15
Ejaaz: their alpha-fold model that is genetically sequencing a bunch of different molecules
17:20
Ejaaz: and creating new kind of drugs.
17:22
Ejaaz: I am really excited personally about this because I have a bit of a science
17:25
Ejaaz: background and I think that,
17:27
Ejaaz: applying these models in ways that can really change human impact,
17:31
Ejaaz: not necessarily for people who are using these models, are going to be huge.
17:34
Josh: I'm like stoked for bio and life sciences and all that improvement.
17:37
Josh: I think we are like very close to getting whatever AGI for science looks like very soon.
17:42
Josh: I think that is going to be a huge category that we're going to see a lot of
17:45
Josh: progress. You see all these labs working towards that.
17:47
Josh: The thing I'm most excited about seeing is how we actually implement all of this intelligence.
17:52
Josh: I think there is this asymmetry happening right now where a few companies are
17:56
Josh: creating an unbelievably capable set of tools that are amazing and powerful.
18:02
Josh: And the world now has to figure out how to extract value from them in their own ways.
18:06
Josh: So when I look at the next 200 episodes, the thing I will be most closely watching
18:11
Josh: is how the world's actually able
18:13
Josh: to extract value from this new form of intelligence that we've created.
18:17
Josh: Because there is this asymmetry where people are really afraid of job loss and
18:20
Josh: people are really afraid of margins getting affected.
18:22
Josh: And the reality is, is that most people, most companies haven't actually implemented this at all.
18:27
Josh: If anything, like the real broad sweeping effects of this across the world have
18:32
Josh: been so small on a relative basis to how powerful they've gotten that I think
18:36
Josh: this dispersion over the next 200 episodes, maybe 14 months,
18:40
Josh: or actually going quicker now.
18:41
Josh: So maybe it'll take like 10 to 11 months to get the next 200 episodes.
18:45
Josh: But the dispersion of this intelligence into day-to-day devices,
18:49
Josh: into day-to-day companies, kind of reimagining what the average role at a company
18:54
Josh: looks like using these new tools.
18:56
Josh: That's what I'm particularly excited about. And then second to that is this energy problem, man.
19:00
Josh: Whoever is going to solve this energy problem, who's ever going to be able to
19:03
Josh: build and create more energy is going to be
19:06
Josh: One of the most powerful companies on earth. We just have such a shortage of it regardless of AI.
19:10
Josh: So I think when I think about whether or not we're in a bubble or where we are
19:14
Josh: in the market cycle, I think those two things are kind of separate from whether
19:18
Josh: or not memory stocks continue to go up, from whether or not NVIDIA continues to make better chips.
19:22
Josh: It's how do we take what we have today and disperse it effectively throughout the world?
19:26
Josh: And then how do we power all this new technology that we are working on creating?
19:29
Josh: And those two problems are the ones that I am most excited about i'm most bullish on i am
19:34
Josh: i guess most interested in and considering like when i think about this
19:37
Josh: the implementation of it that comes in the form of robots we have atoms the
19:41
Josh: new company from travis kalanick who is
19:42
Josh: fully all in on getting precious metals turning those precious metals into robots
19:47
Josh: that are now our use case and then using those robots to distribute materials
19:50
Josh: throughout the world that is so cool we have tesla in the optimist robot program
19:53
Josh: who's focused on building humanoid robots as a form factor and doing that at a scale that we saw
19:58
Josh: for the Model 3 and Model Y, which is now the most popular and highest sold car in the world.
20:02
Josh: So companies that are working on this, I am maximally bullish on and maximally
20:06
Josh: excited about. And I think, yeah, over the next year, those are the things that
20:08
Josh: I am very excited and looking out for.
20:10
Ejaaz: Yeah, the magic is already here. It just hasn't been distributed well enough at this point.
20:15
Ejaaz: When I think about key trends that are really tiny right now,
20:19
Ejaaz: but are going to play out over the next, let's call it over the next 100 episodes,
20:23
Ejaaz: maybe not 200, too ambitious. I don't know if I can predict that far ahead,
20:28
Ejaaz: I think open source models, or let me correct myself, locally run models on
20:33
Ejaaz: edge compute, smaller models that are hyper-optimized to serve you or a particular
20:37
Ejaaz: use case, are going to become way, way, way more abundant.
20:41
Ejaaz: And we're seeing some prime evidence of this happening right now.
20:45
Ejaaz: We've seen big companies like Uber and the likes of Meta cut back on their AI
20:51
Ejaaz: token spending and usage. Why?
20:53
Ejaaz: Because it's quite expensive to use frontier models right now.
20:55
Ejaaz: Now, Fable 5, we were talking about it before we started the show,
20:58
Ejaaz: is $10 in and $50 out per million token input and outputs.
21:03
Ejaaz: And that's expensive if you are burning through millions and millions and millions
21:06
Ejaaz: of these tokens every single week.
21:08
Ejaaz: I think people are going to opt for more locally run models,
21:11
Ejaaz: whether these are Chinese open source models, whether they're American open
21:13
Ejaaz: source models, or in Apple's case, they kind of train and fine tune their own
21:18
Ejaaz: models that they run on their own local hardware.
21:20
Ejaaz: I think we're going to see an acceleration and push towards new types of devices.
21:24
Ejaaz: We're talking about open AI, releasing new devices by the end of the year.
21:27
Ejaaz: We know Apple's working on that. We know a bunch of other companies are working
21:30
Ejaaz: on their own versions of lenses, or whether that might be a pendant that sits
21:33
Ejaaz: on your chest or whatever that might be.
21:36
Ejaaz: And I think we're going to have a hyper-optimized world around smaller models
21:40
Ejaaz: and hardware that listens and sees and hears everything that you do and see.
21:45
Ejaaz: And the point of all of this is personal AI agents, which brings me to my second
21:48
Ejaaz: point, which is, I think we're going to see a lot of people prompting less and
21:53
Ejaaz: relying on AI to do the thing for them and figure out the problem for them.
21:57
Ejaaz: There's this theme that we've spoken about a number of times over the last couple
22:00
Ejaaz: of months, which is RSI, recursive self-improvement.
22:03
Ejaaz: This is basically when an AI model is smart enough to know how to build the
22:07
Ejaaz: next version of itself. Now, this is a really hard problem to crack,
22:10
Ejaaz: but a bunch of really intelligent people have predicted that this is probably
22:14
Ejaaz: going to come at the start of 2027, which is not too long away from here.
22:17
Ejaaz: In the event that we have that that could assume that.
22:21
Ejaaz: AI models can basically run 24 seven and do work for you whilst you sleep or
22:25
Ejaaz: work on any of your problems, whatever that might be.
22:27
Ejaaz: And I think in that world, we are going to increasingly rely on this AI to know
22:32
Ejaaz: more about us, share more personal data with them. And it's kind of a,
22:37
Ejaaz: a concern that I have, I guess, which is like, we're putting all our brainpower
22:40
Ejaaz: into these AI agents and these AI models. I wonder how that kind of like maps
22:43
Ejaaz: out going forwards. We don't know.
22:45
Ejaaz: And then the third final trend that I think is going to play out is,
22:48
Ejaaz: I think China is going to close source a bunch of their models coming soon.
22:52
Ejaaz: There was news that broke this morning that the Chinese government has ordered
22:55
Ejaaz: a bunch of their AI labs to basically cut off access, basically their version
22:59
Ejaaz: of export controls, because they don't want,
23:01
Ejaaz: the West or Americans in particular to get access to their models and so they
23:05
Ejaaz: might move to pricing or they might just go completely private i don't think
23:09
Ejaaz: this is for the benefit of the world
23:11
Ejaaz: but it is increasingly become a geopolitical risk that i think will end up playing
23:15
Ejaaz: out at least over the next six to twelve months.
23:17
Josh: The one certainty is that the world's 100 episodes from now will look absolutely
23:21
Josh: nothing like that today at least as it relates to the world of ai because we
23:25
Josh: are on this accelerating curve and i think one of the things that we mentioned
23:29
Josh: early on, we even had the guests on the show who wrote it was the AGI
23:32
Josh: 2027 paper, where it kind of projected the trajectories and timelines of how
23:37
Josh: fast things were going to go.
23:38
Josh: And so far, to their credit, a lot of that has actually happened in 100 episodes
23:42
Josh: from now, brings us to the middle of 2027.
23:45
Josh: And thinking about life in 2027, a year from now, like you talk about local
23:50
Josh: models, EJs, we're going to be running mythos class models on our cell phones.
23:53
Josh: And like, what are the downstream implications of that?
23:56
Josh: What does the world look like when you truly have that amount of intelligence
23:58
Josh: available at your fingertips on a mobile device? Does it matter that China is
24:03
Josh: closed source because the models will be so powerful that most people don't
24:06
Josh: actually need the upper echelon in order to accomplish their day-to-day tasks?
24:10
Josh: And it's funny, as we're doing these episodes, it feels like because we're living
24:13
Josh: in it, it doesn't progress that fast. But looking back at episode one to now,
24:17
Josh: like, oh my God, the world is so different.
24:19
Josh: And looking forward, now that we are aware that we do have the idea of like
24:23
Josh: recursive self-improvement, that things can actually get perpetually better.
24:26
Josh: And as they improve, they continue to improve themselves. And we get this machine
24:30
Josh: that creates this vertical line of trajectory. I think it's gonna look...
24:33
Josh: Pretty weird. And I hope that we're able to convert that into something useful.
24:38
Josh: It's like, okay, we have all this intelligence. What do we do with it?
24:41
Josh: How can we actually implement it throughout the world? What types of robotic
24:44
Josh: figures can we put that into? What types of companies can we build because of it?
24:47
Josh: How can you empower employees to do amazing things with it? It's like,
24:50
Josh: these are the questions that I'm really excited to answer and to find the companies
24:53
Josh: who are doing the best at answering those questions in particular.
24:57
Josh: And then they're like companies, I'm thinking about Etched. I'm like,
24:59
Josh: oh my God, if they can create the same amount of tokens for a tenth of the cost like
25:04
Josh: what impact does that have on the world so there's a lot of really fun and crazy
25:07
Josh: and exciting things coming down the pipe and we are going to continue along publishing
25:13
Josh: four times a week we've been cooking um and yeah i think that's that's mostly
25:17
Josh: the episode here is uh episode 200 dude what a journey it's been that's uh
25:21
Josh: That's a lot of times click and record.
25:23
Ejaaz: We were mere younglings at the start of this, Josh. We were doing two to three episodes a week.
25:29
Ejaaz: We were still learning about a lot of the concepts that we speak about quite
25:32
Ejaaz: frequently on the show today. We've come a long way, dude.
25:36
Ejaaz: And I'm feeling pretty optimistic. And I guess that's the final point that I
25:39
Ejaaz: want to mention is I'm unbearably optimistic about what the future is.
25:44
Ejaaz: In fact, I have a stat for you that I could share up here, which is the amount
25:51
Ejaaz: of times we have mentioned bullish over the last 200 episodes,
25:55
Ejaaz: 192 times versus bearish, just 68.
25:59
Ejaaz: So the point is, Josh and myself are incredibly optimistic about the future.
26:03
Ejaaz: I personally think that AI tools right now, no matter how scary,
26:07
Ejaaz: intelligent they get, is going to result in an abundance or an explosion of.
26:12
Ejaaz: Advancements in GDP, in advancements in science and new cures.
26:17
Ejaaz: And the best part is I think it isn't going to be age-gapped.
26:21
Ejaaz: I think you could be a 24-year-old that drops out of Harvard and builds the
26:24
Ejaaz: most amazing chip that competes directly with NVIDIA.
26:27
Ejaaz: Or you could be someone that is just learning science, AP science at school.
26:31
Ejaaz: And you might use a model because you have a certain idea and you discover something
26:34
Ejaaz: that NASA ends up sponsoring. That is exactly what happened to their current
26:37
Ejaaz: intern that is working there right now.
26:39
Ejaaz: So there's going to be an abundance and explosion of intelligence that is available
26:44
Ejaaz: to anyone and everyone that can get access to these models.
26:46
Ejaaz: And I think whether it's open source, whether it's private, whether it's a $20
26:49
Ejaaz: a month membership, you will be able to participate in this optimistic future.
26:55
Ejaaz: And I'm very, very excited about that potential fact.
26:58
Josh: You can't just scroll to the section and not go through the rest.
27:00
Josh: Bullish versus bearish index 2.8 to one.
27:03
Josh: That's so funny. The absolutely insane index. EJAS has captured 23 of 28 mentions. It is insane.
27:11
Josh: It has a subsection. It says ownership established beyond statistical doubt.
27:14
Josh: And then for the exponential, I have captured 48 of those exponentials to 28.
27:19
Josh: It says the house physicist has a favorite curve, which is pretty good.
27:24
Josh: And then this is this is the set that i'm calling the the waffle index um and
27:28
Josh: it's 38.9 versus 27.9 hours talks it says nearly identical turn counts one host
27:34
Josh: turns run 40 longer the database knows who he
27:37
Ejaaz: Just i think i'm sorry can i interrupt you who who might that be i don't know
27:41
Ejaaz: um i don't really talk that much but uh yeah no i don't take i don't take up
27:45
Ejaaz: the mic space that much i guess i'd like to think listen it's value it's as
27:49
Ejaaz: much value as i can and also it's like.
27:50
Josh: I just like that we could quantify it now exactly
27:53
Ejaaz: My own recursive self-learning process that i have to subject all our listeners to what a.
27:58
Josh: Journey what a freaking journey so that's 200 episodes in the books for anyone who's been here
28:03
Josh: even like for the last 50 thank you for the ones who've been here since day
28:06
Josh: one for the ones who have come over and joined us and been along this journey
28:11
Josh: to support us like thank you this is pretty crazy the podcast mostly started
28:15
Josh: as an accident uh david from bankless he was like hey do you want to like do
28:19
Josh: your own podcast and I was like yeah I guess so and
28:23
Josh: I said that and then I went away on a vacation for a week and I came back and
28:26
Josh: we had a podcast and I was like okay I guess we're doing this and we did it and
28:29
Josh: people really supported and they enjoyed it. And I think we've done some pretty
28:33
Josh: amazing work so far. So thank you to everyone who's been a part of the journey.
28:37
Josh: It's really been special.
28:39
Josh: Like when I look back on everything, I'll be like, wow, that was a really cool
28:42
Josh: part of life that we got to do and creating this amazing show.
28:46
Josh: And it still feels very early days. There's so much left to cover.
28:49
Josh: There's so much momentum still happening. There's so many special things worth
28:53
Josh: talking about in the world of AI, but also just
28:56
Josh: on the frontier of technology, which is originally the idea of limitless it's
28:59
Josh: like it is this is exploring the frontier of all the things ai continues to
29:04
Josh: be that hottest thing but there's so much more coming down the line that downstream
29:07
Josh: of ai that i'm very excited to talk about and cover so
29:10
Josh: that is that's the journey that's 200 for the real ones still listening for
29:14
Josh: the real ones who've been watching forever like oh my god you're absolute legends
29:18
Josh: thank you so much for being here it's like the only reason why this is possible
29:21
Josh: the only reason why i've been able to do this for so long is because all of
29:24
Josh: you um showing love and support every single episode.
29:26
Josh: So sincerely, thank you so much for all of the support over this time.
29:31
Ejaaz: Thank you, guys. And if you are listening to us right now and you aren't subscribed,
29:35
Ejaaz: or if you haven't commented, or if you haven't given us a thumbs up ever, try it out today.
29:39
Ejaaz: What better day to do it than on our 200th episode? It would be awesome to hear from you guys.
29:44
Josh: Also, wait, we got to talk about our stats too. It's like, guys,
29:47
Josh: we're a top 30 podcast in the world in technology. Like, this isn't just a look like...
29:51
Ejaaz: No, open up Spotify. Open up the charts right now. Yeah, the boys are cooking.
29:55
Josh: So I think that's not only have we done it for 200 episodes but you have given
29:58
Josh: us enough support to put us on the top of the charts which is freaking awesome
30:01
Josh: so that is really exciting that is really reassuring all good things to come
30:05
Josh: any final parting thoughts before we wrap this up get ready for 201 thank you guys
30:09
Ejaaz: For joining us on this journey and we will be here hopefully for the next 200
30:14
Ejaaz: we are excited to bring you all the news and insights and trends that we can
30:18
Ejaaz: possibly fathom even though I speak 40% longer it'll be worth it and we're excited
30:23
Ejaaz: to see you very soon And we'll see.
30:24
Josh: You on the next episode. Cheers to 200 more.
30:26
Ejaaz: Cheers to 200 more. See you guys.