Unsupervised Learning x Latent Space Crossover Special

If you’re in SF: Join us for the Claude Plays Pokemon hackathon this Sunday!

If you’re not: Fill out the 2025 State of AI Eng survey for $250 in Amazon cards!

Unsupervised Learning is a podcast that interviews the sharpest minds in AI about what’s real today, what will be real in the future and what it means for businesses and the world - helping builders, researchers and founders deconstruct and understand the biggest breakthroughs.

Top guests: Noam Shazeer, Bob McGrew, Noam Brown, Dylan Patel, Percy Liang, David Luan

Full Episode on Their YouTube

Timestamps

* 00:00 Introduction and Excitement for Collaboration

* 00:27 Reflecting on Surprises in AI Over the Past Year

* 01:44 Open Source Models and Their Adoption

* 06:01 The Rise of GPT Wrappers

* 06:55 AI Builders and Low-Code Platforms

* 09:35 Overhyped and Underhyped AI Trends

* 22:17 Product Market Fit in AI

* 28:23 Google's Current Momentum

* 28:33 Customer Support and AI

* 29:54 AI's Impact on Cost and Growth

* 31:05 Voice AI and Scheduling

* 32:59 Emerging AI Applications

* 34:12 Education and AI

* 36:34 Defensibility in AI Applications

* 40:10 Infrastructure and AI

* 47:08 Challenges and Future of AI

* 52:15 Quick Fire Round and Closing Remarks

Transcript

[00:00:00] Introduction and Podcast Overview

[00:00:00] Jacob: well, thanks so much for doing this, guys. I feel like we've we've been excited to do a collab for a while. I

[00:00:13] swyx: love crossovers. Yeah. Yeah. This, this is great. Like the ultimate meta about just podcasters talking to other podcasters. Yeah. It's a lot. Podcasts all the way up.

[00:00:21] Jacob: I figured we'd have a pretty free ranging conversation today but brought a few conversation starters to, to, to kick us off.

[00:00:27] Reflecting on AI Surprises and Trends

[00:00:27] Jacob: And so I figured one interesting place to start is you know, obviously it feels that this world is changing like every few months. Wondering as you guys reflect path on the past year, like what surprised you the most?

[00:00:36] Alessio: I think definitely recently models we kinda on the, on the right here. Like, oh, that, well, I, I I think there's, there's like the, what surprised us in a good way.

[00:00:44] May maybe in a, in a bad way. I would say in a good way. Recently models and I think the release of them right after the new reps scaling instead talked by Ilia. I think there was maybe like a, a little. It's so over and then we're so back. I'm like such a short, short period. It was really [00:01:00] fortuitous

[00:01:00] Jacob: timing though, like right.

[00:01:01] As pre-training died, I mean, obviously I'm sure within the labs they knew pre-training was dying and had to find something. But you know, from the outside it was it, it felt like one right into the other.

[00:01:09] Alessio: Yeah. Yeah, exactly. So that, that was a good surprise,

[00:01:12] swyx: I would say, if you wanna make that comment about timing, I think it's suspiciously neat that like, because we know that Strawberry was being worked on for like two years-ish.

[00:01:20] Like, and we know exactly when Nome joined OpenAI, and that was obviously a big strategic bet by OpenAI. So like, for it to transition, so transition so nicely when like, pre-training is kind of tapped out to, into like, oh, now inference time is, is the new scaling law is like conv very convenient. I, I, I like if there were an Illuminati, this would be what they planned.

[00:01:41] Or if we're living in a simulation or something. Yeah.

[00:01:44] Open Source Models and Their Impact

[00:01:44] swyx: Then you said open source

[00:01:45] Alessio: as well? Yeah. Well, no, I, I think like open source. Yeah. We're discussing this on the negative. I would say the relevance of open source. I would specifically open models. Yeah, I was surprised the lack, like the llamas of the world by the lack of adoption.

[00:01:56] And I mean, people use it obviously, but I would say nobody's [00:02:00] really like a huge fanboy, you know, I think the local llama community and some of the more obvious use cases really like it. But when we talk to like enterprise folks, it's like, it's cool, you know? And I think people love to argue about licenses and all of that, but the reality is that it doesn't really change the adoption path of, of ai.

[00:02:18] So

[00:02:19] swyx: yeah, the specific stat that I got from on anchor from Braintrust mm-hmm. In one of the episodes that we did was I think he estimated that open source model usage in work in enterprises is that like 5% and going down.

[00:02:31] Jacob: And it feels like you're basically all these enterprises are in like use case discovery mode, where it's like, let's just take what we think is the most powerful model and figure out if we can find anything that works.

[00:02:39] And, you know, so much of, of, of it feels like discovery of that. And then, right, as you've discovered something, a new generation of models are out and so you have to go do discovery with those. And you know, I think obviously we're probably optimistic that the that the open source models increase in uptake.

[00:02:50] It's funny, I was gonna say my biggest surprise in the last year was open source related, but it was just how Fast Open Source caught up on the reasoning models. It was kind of unclear to me, like over time whether there would be, you know, [00:03:00] a compounding advantage for some of the closed source models where in the, okay, in the early days of, of scaling you know, there was a, a tight time loop, but over time, you know, would would the gap increase?

[00:03:08] And if anything it feels like a trunk. You know, and I think deep seek specifically was just really surprising in how, you know, in many ways if the value of these model companies is like you have a model for a period of time and you're the only one that can build products on top of that model while you have it.

[00:03:21] Like, God, that time period is a lot shorter than a, than I thought it was gonna be a year ago.

[00:03:25] swyx: Yeah. I mean, again, I I, I don't like this label of how Fast Open Source caught up because it's really how Fast Deepsea caught up. Right. And now we have, like, I think some of it is that Deepsea is basically gonna stop open sourcing models.

[00:03:36] Yeah. So like there, there's no team open source, there's just different companies and they choose to open source or not. And we got lucky with deep seek releasing something and then everyone else is basically distilling from deep seek and those are distillations. Catching up is such an easier lower bar than like actually catching up, which is like you, you are like from scratch.

[00:03:56] You're training something that like is competitive on that front. I don't know if [00:04:00] that's happening. Like basically the only player right now is we're waiting for LA four.

[00:04:03] Jordan: I mean, it's always an order of magnitude cheaper to replicate what's already been done than to create something fundamentally new.

[00:04:09] And so that's why I think deep seek overall was overhyped. Right? I mean obviously it's a good open source, new entrant, but at the same time there's nothing new fundamentally there other than sort of doing it executing what's already been done really well.

[00:04:21] Alessio: Yeah,

[00:04:21] Jordan: right.

[00:04:21] Alessio: So Well, but I think the traces is like maybe the biggest thing, I think most previous open models is like the same model, just a little worse and cheaper.

[00:04:30] Yeah. Like R one is like the first model that had the full traces. So I think that's like a net unique thing in fair, open source. But yeah, I, I think like we talked about deep seek in the our n of year 2023 recap, and we're mostly focused on cheaper inference. Like we didn't really have deep, see, deep CV three

[00:04:47] swyx: was out then, and we were like, that was already like talking about fine green mixture of experts and all that.

[00:04:51] Like that's a great receipt to

[00:04:52] Jacob: have

[00:04:52] swyx: to be like, yeah.

[00:04:52] Jacob: End

[00:04:53] swyx: of year 20. Yeah. That's a,

[00:04:54] Jacob: that's a, that's, that's an

[00:04:55] swyx: impressive one. You follow the right whale believers in Twitter. It's, it's like [00:05:00] pretty obvious. I actually had like so, you know, I used to be in finance and, and a lot, a lot of my hedge fund and PE friends called me up.

[00:05:06] They were like, why didn't you tip us off on deep seek? And I'm like, well, I mean, it's been there. It's, it's actually like kind of surprising that like, Nvidia like fell like what, 15% in one day? Yeah. Because deep seek and I, I think it's just like whatever the market, public market narrative decides is a story, becomes the story, but really like the technical movements are usually.

[00:05:26] One to two years in the making. Before that,

[00:05:27] Jacob: basically these people were telling on themselves that they didn't listen to your podcast. They've been on the end of year 22, 3. No, no,

[00:05:32] swyx: no. Like yeah, we weren't, we weren't like banging the drum. So like it's also on us to be like, no, like this. This is an actual tipping point.

[00:05:38] And I think I like as people who are like, our function as podcasters and industry analysts is to raise the bar or focus attention on things that you think matter. And sometimes we're too passive about it. And I think I was too passive there. I'd be, I'd be happy to own up on that.

[00:05:52] Jacob: No, I feel like over time you guys have moved into this margin general role of like taking stances of things that are or aren't important and, you know I feel like you've done that with MCP of [00:06:00] late and a bunch of

[00:06:00] swyx: things.

[00:06:00] Yeah.

[00:06:01] Challenges and Opportunities in AI Engineering

[00:06:01] swyx: So like the, the general pushes is AI engineering, you know, like it's gotta, gotta wrap the shirt. And MCP is part of that, but like the, the general movement is what can engineers do above the model layer to augment model capabilities. And it turns out it's a lot. And turns out we went from like, making fun of GPT rappers to now I think the overwhelming consensus GPT wrappers is the only thing that's interesting.

[00:06:20] Yeah.

[00:06:21] Jacob: I remember like, Arvin from Perplexity came on our podcast and he was like, I'm proudly a rapper. Like, you know, it's like anyone that's like talking about like, you know, differentiation, like pre-product market fit is like a ridiculous thing to, to say, like, build something people want and then yeah.

[00:06:33] Over time you can kind of worry about that.

[00:06:35] swyx: Yeah. I, I interviewed him in 2023 and I think he may have been the first person on our podcast to like, probably be a GBT rapper. Yeah. And yeah, and obviously he's built a huge business on that. Totally. Now, now we now we all can't get enough of it. I have another one for, Oh, nice.

[00:06:47] That was Alessia's one and we, we perhaps individual answers just to be interesting in the same Uber on the way up. Yeah. You just like in the, in different Oh, I was driving too. Oh, you were driving. So I actually, I mean, it was a Tesla mostly drove mine was [00:07:00] actually, it is interesting that low-code builders did not capture the AI builder market.

[00:07:04] Right. AI builders being bought lovable, low-code builders being Zapier, Airtable, retool notion. Any of those, like you're not technical. You can build software.

[00:07:14] misc: Yeah.

[00:07:14] swyx: Somehow not all them missed it. Why? It's bizarre. Like they should have the DNA, I don't know. They should have. They already have the reach, they already have the, the distribution.

[00:07:25] Like why? I I have no idea. The ability to

[00:07:27] Jacob: fast follow too. Like I'm surprised there's Yeah. There's just

[00:07:29] swyx: nothing. Yeah. What do you make of that? I, it seems and you know, not to come back to the AI engineering future, like it takes a, a certain kind of. Founder mindset or AI engineer mindset to be like, we will build this from whole cloth and not be tied to existing paradigms.

[00:07:45] I think, 'cause I like, if I was, if I'm to, you know, you know, Wade or who's, who's, who's the Zapier person than, you know, Mike. Mike who has left the Zapier. Yeah. What's the, yeah. Like you know, Zapier, when they decided to do Zapier ai, they [00:08:00] were like, oh, you can use natural language to make Zap actions, right?

[00:08:03] When Notion decided to do Notion ai, they were like, oh, you can like, you know write documents or, you know, fill in tables with, with ai. Like, they didn't do the, the, the, the next step because they already had their base and they were like, let's improve our baseline. And the other people who actually tried for to, to create a phone cloth were like, we, we got no prior preconceptions.

[00:08:24] Like, let's see what we can, what kinda software people can build with like from scratch, basically. I don't know that, that's my explanation. I dunno if you guys have any retros on the AI builders?

[00:08:33] Jacob: Yeah. Or, or, or did they kind of get lucky getting, you know starting that product journey? Like right as the models were reaching the inflection point?

[00:08:39] There's the timing

[00:08:40] swyx: issue. Yeah. Yeah, yeah. Yeah. Yeah, I don't know. Like I, I, to some extent, I think the only reason you and I are talking about it is that they, both of them have reported like ridiculous numbers. Like zero to 20 million in three months, basically, both of them. Jordan, did you have a, a big surprise?

[00:08:55] Jordan: Yeah, I mean, some of what's already been discussed. I guess the only other thing would be on the Apple side in particular, I [00:09:00] think, I think you know, for the last text message summary, like, but they're

[00:09:04] Jacob: funny. They're funny at how bad they had, how off they're, they're viral. Yeah.

[00:09:08] Jordan: I mean, so like for the last couple years we've seen so many companies that are trying to do personal assistance, like all these various consumer things, and one of the things we've always asked is, well, apple is in prime position to do all this.

[00:09:18] And then with Apple Intelligence, they just. Totally messed up in so many different ways. And then the whole BBC thing saying that the guy shot himself when he didn't. And just like, there's just so many things at this point that I would've thought that they would've ironed up their, their AI products better, but just didn't really catch on,

[00:09:35] Jacob: you know, second on this list of, of generally overly broad opening questions would be anything that you guys think is kind of like overhyped or under hyped in the AI world right now?

[00:09:43] Alessio: Overhyped agents framework. Sorry. Not naming any particular ones. I'm sorry. Not, not not, yeah, exactly. It's not, I, I would say they're just overall a chase to try and be the framework when the workloads are like in such flux. Yeah. That I just think is like so [00:10:00] hard to reconcile the two. I think what Harrison and Link Chain has done so amazingly, it's like product velocity.

[00:10:05] Like, you know, the initial obstructions were maybe not the ending obstruction, but like they were just releasing stuff every day trying to be on top of it. But I think now we're like past that, like what people are looking for now. It's like something that they can actually build on mm-hmm. And stay on for the next couple of years.

[00:10:23] And we talked about this with Brett Taylor on our episode, and it feels like, it's like the jQuery era Yeah. Of like agents and lms. It's like, it's kinda like, you know, single file, big frameworks, kinda like a lot of players, but maybe we need React. And I think people are just trying to build still Jake Barry.

[00:10:39] Like, I don't really see a lot of people doing react like,

[00:10:43] swyx: yeah. Maybe the, the only modification I made about that is maybe it's too early even for frameworks at all. And the thing that, and do you think

[00:10:50] Jacob: there's enough stability in the underlying model layer and, and patterns to, to have this,

[00:10:54] swyx: the thing is the protocol and not the framework?

[00:10:56] Jacob: Yeah.

[00:10:56] swyx: Because frameworks inherently embed protocols, but if you just focus on a protocol, maybe that [00:11:00] works. And obviously MCP is. The current leading mm-hmm. Area. And you know, I think the comparison there would be, instead of just jQuery, it is XML HTB requests, which is like the, the thing that enabled Ajax.

[00:11:10] And that was the, the, the, the, the sort of inciting incident for JavaScripts being popular as a language.

[00:11:16] Jordan: I would largely agree with that. I mean, I think on the, the react side of things, I think we're starting to see more frameworks sort of go after more of that, I guess like master is sort of like on the TypeScript side and more of like a sort of master.

[00:11:28] Yeah, yeah, yeah, yeah. The traction is really impressive there. And so I think we're starting to see more surface there, but I think there's still a big opportunity. What do you have for for an over or under hyped on the under hype side? You know, I actually, I, I know I mentioned Apple already, but I think the private cloud compute side with PCC, I actually think that could be really big.

[00:11:45] It's under the radar right now. Mm-hmm. But in terms of basically bringing. The on device sort of security to the cloud. They've done a lot of architecturally interesting things there. Who's they? Apple. Oh, okay. On the PCC side. And so I actually think of that.

[00:11:58] swyx: So you're negative on Apple [00:12:00] Intelligence, but also on Apple Cloud,

[00:12:01] Jordan: on the more of the local device.

[00:12:04] Sort of, I think there'll be a lot of workloads still on device, but when you need to speak to the cloud for larger LLMs, I think that Apple has done really interesting thing on the privacy side.

[00:12:13] Alessio: Yeah. We did the seed of a company that does that, so Yeah. Especially as things become more co that you set 'em up on purpose.

[00:12:18] So that felt like a perfect Yeah, no, I was like, let's go Jordan, you guys concluding before this episode? Tell me about that company after. We'll chat after, but, but yes, I, I think that's like the unique the thing about LLM workflows is like you just cannot have everything be single tenant, right?

[00:12:35] Because you just cannot get enough GPUs. Like even like large enterprises are used to having VPCs and like everything runs privately. But now you just cannot get enough GPUs to run in a VPC. So I think you're gonna need to be in a multi-tenant architecture, and you need, like you said, like single tenant guarantees in multi-tenant environment.

[00:12:52] So yeah, it's a interesting space.

[00:12:55] swyx: Yeah. What about you, Swiss? Under hypes, I want to say [00:13:00] memory. Just like stateful ai. As part of my keynote on, on for just like every, every conference I do, I do a keynote and I try to do the task of like defining an agent, just, you know, always evergreen content, every content for a keynote.

[00:13:14] But I did it in a, in a way that it was like I think like a, what a researcher would do. Like you, you survey what people say and then you sort of categorize and, and go like, okay, this is the, the. What everyone calls agents and here are the groups of DEF definitions. Pick and choose. Right. And then it was very interesting that the week after that OpenAI launched their agents SDK and kind of formalized what they think agents are.

[00:13:34] CloudFlare also did the same with us and none of them had memory. Yeah, it's very strange. The, pretty much like the only big lab o obviously there, there's conversation memory, but there's not memory memory like in like a, like a let's store a large across fact about you and like, you know, exceed the, the context length.

[00:13:54] And here's the, if you, if you're look, if you look closely enough, there's a really good implementation of memory inside of [00:14:00] MCP when they launched with the initial set of servers. They had a memory server in there, which I, I would recommend as like, that's where you start with memory. But I think like if there was a better, I.

[00:14:10] Memory abstraction, then a lot of our agents would be smarter and could learn on, on the job, which is something that we all want. And for some reason we all just like ignored that because it's just convenient to, and, but do you feel like

[00:14:24] Jacob: it's being ignored or it's just a really hard problem and like lots of, I feel like lots of people are working on it.

[00:14:27] Just feels like it's, it's proven more challenging.

[00:14:29] swyx: Yeah. Yeah. Yeah. So, so Harrison has lang me, which I think now he's like, you know, relaunched again. And then we had letter come speak at our mm-hmm. Our conference I don't know, Zep, I think there's a bunch of other memory guys, but like, something like this I think should be normal in the stack.

[00:14:44] And basically I think anything stateful should be interesting to VCs 'cause it's databases and, you know, we know how those things make money.

[00:14:51] Jacob: I think on the over hype side, the only thing I'd add is like, I'm, I'm still surprised how many net new companies there are training models. I thought we were kind of like past that.

[00:14:58] And

[00:14:58] swyx: I would say they died end of last year. And now, [00:15:00] now they've resurfaced. Yeah. I mean they, that's one of the questions that you had down there of like, yeah. Sorry. Is there an opportunity for net new model players? I wouldn't say no. I don't know what you guys think.

[00:15:08] Alessio: I, I don't have a reason to say no, but I also don't have a reason to say, this is what is missing and you should have a new model company do it.

[00:15:15] But again, I'm an add here. Like, all these guys wanna

[00:15:17] swyx: pursue a GI, you know, all, they all want to be like, oh, we'll, we'll like hit, you know, soda on all the benchmarks and like, they can't all do it. Yeah.

[00:15:25] Jacob: I mean, look, I don't know if Ilia has the secret secret approach up his sleeve of of something beyond test time compute.

[00:15:29] Mm-hmm. But it was funny, I, we had Noam Shaer on the podcast last week. I was asking him like, you know, is, is there like some sort of other algorithmic breakthrough? Would he make a Ilia? And he's like, look, I think what he is implicitly said was test time compute gets to the point where these models are doing AI engineering for us.

[00:15:43] And so, you know, at that point they'll figure out the next algorithm breakthrough. Yeah. Which I thought was was pretty interesting.

[00:15:47] Jordan: I agree with you folks. I think that we're most interested, at least from our side and like, you know, foundation models for specific use cases and more specialized use cases.

[00:15:55] Mm-hmm. I guess the broader point is if there is something like that, that these companies can latch onto [00:16:00] and being there sort of. Known for being the best at. Maybe there's a case for that. Largely though I do agree with you that I don't think there should be, at this point, more model companies. I think it's like

[00:16:09] Jacob: these

[00:16:09] Jordan: unique data

[00:16:09] Jacob: sets, right?

[00:16:10] I mean, obviously robotics has been an area we've been really interested in. It's entirely different set of data that's required, you know, on top of like a, a good BLM and then, you know, biology, material sciences, more the specific use cases basically. Yeah. But also specific, like specific markets. A lot of these models are super generalizable, but like, you know finding opportunities to, you know, where, you know, for a lot of these bio companies, they have wet labs, like they're like running a ton of experiments or you know, same on the material sciences side.

[00:16:31] And so I still feel like there's some, some opportunities there, but the core kind of like LLM agent space is it's tough, tough to compete with the big ones.

[00:16:38] Alessio: Yeah. Agree. Yeah. But they're moving more into product. Yeah. So I think that's the question is like, if they could do better vertical models, why not do that instead of trying to do deep research and operator?

[00:16:50] And these different things. Mm-hmm. I think that's what I'm, in my mind, it's like the agents coming

[00:16:53] swyx: out too.

[00:16:54] Alessio: Well. Yeah. In my, in my mind it's like financial pressure. Like they need to monetize in a much shorter timeframe [00:17:00] because the costs are so high. But maybe it's like, it's not that easy to, do

[00:17:04] Jacob: you think they would be, that it would be a better business model to like, do a bunch of vertical?

[00:17:07] Well, it's more like

[00:17:07] Alessio: why wouldn't they, you know, like you make less enemies if you're like a model builder, right? Yeah. Like, like now with deep research and like search, now perplexity like an enemy and like a, you know, Gemini deep research is like more of an enemy. Versus if they were doing a finance model, you know?

[00:17:25] Mm-hmm. Or whatever, like they would just enable so many more companies and they always have, like they had as one of the customer case studies for GBT search, but they're not building a finance based model for them. So is it because it's super hard and somebody should do it? Or is it because the new models.

[00:17:41] Are gonna be so much better that like the vertical models are useless anyways. Like this is better lesson. Exactly.

[00:17:46] Jacob: It still seems to be a somewhat outstanding question. I, I'd say like, all the signs of the last few years seem to be like a general purpose model is like the way to go. And, you know, you know, like training a hyper-specific model in this, in, in a domain is like, you know, maybe it's cheaper and faster, but it's not gonna be like higher quality.

[00:17:59] But [00:18:00] also like, I think it's still an, I mean, we were talking to, to no and Jack Ray from Google last week, and they were like, yeah, this is still an outstanding, like, we, we check this every time we have a new model. Like whether there's you know, there that still seems to be holding. I remember like a few years ago, it felt like all the rage was like the, it was like the Bloomberg GPT model came out.

[00:18:14] Everyone was like, oh, you gotta like, you know, massive data. Yeah. I had

[00:18:17] swyx: a GPA, I had DP of AI of Bloomberg present on that. Yeah. That must be a really

[00:18:20] Jacob: interesting episode to go back on because I feel like, like very shortly thereafter, the next opening AI model came out and just like beat it on all sorts of

[00:18:25] swyx: No, it, it was a talk.

[00:18:26] We haven't released it yet, but yeah, I mean it's basically they concluded that the, the closed models were better so they just Yeah. Stopped. Interesting. Exactly. So I feel like that's been the but he's I, I would be. He's very insistent that the work that they did, the team he assembled, the data that he collected is actually useful for more than just the model.

[00:18:42] So like, basically everything but the model survived. What are the other things? The data pipeline. Okay. The team that they, they, they assembled for like fine tuning and implementing whatever models they, they ended up picking. Yeah, it seems like they are happy with that. And they're running with that.

[00:18:57] He runs like 12, 13 [00:19:00] teams at Bloomberg just working. Jenny, I across the company.

[00:19:03] Jacob: I mean, I guess we've, we've all kind of been alluding it to it right now, but I guess because it's a natural transition. You know, the other broad opening I have is just what we're paying most attention to right now. And I think back on this, like, you know, the model company's coming into the product area.

[00:19:13] I mean, I think that's gonna be like, I'm fascinated to see how that plays out over the next year and kind of these like frenemy dynamics and it feels like it's gonna first boil up on like cursor anthropic and like the way that plays out over the next six months I think will be. What, what is Cursor?

[00:19:26] swyx: Anthropic is, you mean Cursor versus anthropic or, yeah. And I

[00:19:29] Jacob: assume, you know, over time Anthropic wants to get more into the application side of coding Uhhuh. And you know, I assume over time Cursor will wanna diversify off of, you know, just using the Anthropic model.

[00:19:39] swyx: It's interesting that now Cursor is now worth like 10 billion, nine, nine, 10 billion.

[00:19:43] Yeah. And like they've made themselves hard to acquire, like I would've said, like, you should just get yourself to five, 6 billion and join OpenAI. And like all the training data goes through OpenAI and that's how they train their coding model. Now it's not as complicated. Now they need to be an independent company.

[00:19:57] Jacob: Increasingly, it's seems to the model companies want to get into the [00:20:00] product layer. And so seeing over the next six, 12 months does having the best model, you know let you kind of start from a cold start on the product side and, and get something in market. Or are the, you know, companies with the best products, even if they eventually have to switch to a somewhat worse, tiny bit worse model, does it not, you know, where do the developers ultimately choose to go?

[00:20:16] I think that'll be super interesting. Yeah.

[00:20:18] Alessio: Don't you think that Devon is more in trouble than cursor? I, I feel like on Tropic, if anything wants to move more towards, I don't think they wanna build the ID like if I think about coding, it's like kind of like, you know, you look at it like a cube, it's like the ID is like one way to get the code and then the agent is like the other side.

[00:20:33] Yeah. I feel like on Tropic wants more be on the agent side and then hand you off the cursor when you want to go in depth versus like trying to build the claw. IDEI think that's not, I would say, I don't know how you think the

[00:20:46] swyx: existence, a cloud code doesn't show, doesn't support what you say. Like maybe they would, but

[00:20:52] Jacob: assume, like I assume both just converge eventually where you want have where will you be able to do both?

[00:20:57] So,

[00:20:57] swyx: so in order to be so we're, we're talking [00:21:00] about coding agents, whether it's sort of what is it? Inner loop versus auto loop, right? Like inner loop is inside cursor, inside your ID between inside of a GI commit and auto loop is between GI commits on, on the cloud. And I think like to be an outer loop coding agent, you have to be more of a, like, we will integrate with your code base, we'll sign your whatever.

[00:21:17] You know, security thing that you need to sign. Yeah. That kinda schlep. I don't think the model ads wanna do that schlep, they just want to provide models. So that, that, that's, that would be my argument against like why cognition should still have, have, have some moat against anthropic just simply because they cognition would do the schlep and the biz dev and the infra that philanthropic doesn't really care about.

[00:21:39] Jacob: I know the schlep is pretty sticky though. Once you do it,

[00:21:41] swyx: it's very sticky. Yeah. Yeah. I mean it's, it's, it's interesting. Like, I, I think the natural winner of that should be sourcegraph. But there's another

[00:21:47] Jacob: unprompted point portfolio. Nice. We, I mean they, they're

[00:21:51] swyx: big supporters like very friendly with both Quinn and B and they've they've done a lot of work with Cody, but like, no, not much work on the outer [00:22:00] loop stuff yet.

[00:22:01] But like any company where like they have already had, like, we've been around for 10 years, we, we like have all the enterprise contracts that you already trust us with your code base. Why would you go trust like factory or cognition as like, you know, 2-year-old startups who like just came outta MIT Like, I don't know.

[00:22:17] Product Market Fit in AI

[00:22:17] Jacob: I guess switching gears to the to the application side I'm curious for both of you, like how do you kind of characterize what has genuine product market fit in AI today? And I guess less, you more and your side of the investing side, like more interesting to invest in that category of the stuff that works today or kind of where the capabilities are going long term.

[00:22:35] Alessio: That's hard. I was asking you to do my job for you, like, man, that's a easy, that's a layout. Tell us all your investing

[00:22:40] pieces. Yeah, yeah, yeah. I, I, I would say we, well we only really do mostly seed investing, so it's hard to invest in things that already work. Yeah. That fair. Are really late. So we try to, but, but we try to be at the cusp of like, you know, usually the investments we like to make, there's like really not that much market risk.

[00:22:57] It's like if this works. Obviously people are gonna [00:23:00] use it, but like it's unclear whether or not it's gonna work. So that's kind of more what we skew towards. We try not to chase as many trends and I don't know, I, you know, I was a founder myself and sometimes I feel like it's easy to just jump in and do the thing that is hot, but like becoming a founder to do something that is like underappreciated or like doesn't yet work shows some level of like dread and self, like you, you actually really believe in the thing.

[00:23:25] So that alone for me is like, kind of makes me skew more towards that. And you do a lot of angel investing too, so I'm curious how,

[00:23:31] swyx: Yeah, but I don't regard, I don't have, I don't use, put, put that in my mental framework of things like I come at this much more as a content creator or market analyst of like, yeah, it, it really does matter to me what has part of market fit because.

[00:23:45] People, I have to answer the question of what is working now When, when people ask me,

[00:23:50] Jacob: do you feel like relative to the, the obviously the hype and discourse out there, like, you know, do you feel like there's a lot of things that have product market fit or like a few things, like where a few things? Yeah.

[00:23:58] swyx: I was gonna say this, so I have a list [00:24:00] of like two years ago we, I wrote the Anatomy of autonomy posts where it was like the, the first, like what's going on in agents and, and and, and, and what is actually making money. Because I think there's a lot of gen I skeptics out there. They're all like, these, these things are toys.

[00:24:13] They're, they're not unreliable. And you know, why, why, why you dedicating your life to these things. And I think for me, the party market fit bar at the time was a hundred million dollars, right? Like what use cases can reasonably fit a hundred million dollars. And at the time it was like co-pilot it was Jasper.

[00:24:30] No longer, but mm-hmm. You know, in that category of like help you write. Yeah. Which I think, I think was, was helpful. And then and the cursor I think was on there as, as a, as, as, as like a coding agent. Plus plus. I think that list will just grow over time of like the form factors that we know to work, and then we can just adapt the form factors to a bunch of other things.

[00:24:47] So like the, the one that's the most recently added to this is deep research.

[00:24:52] misc: Yeah.

[00:24:52] swyx: Right. Where anything that looks like a deep research whether it's a grok version, Gemini version, perplexity version, whatever. He has an investment [00:25:00] that that he likes called Brightwave that is basically deep research for finance.

[00:25:02] Yeah. And anything where like all it is like long-term agent, agent reporting and it's starting to take more and more of the job away from you and, and just give you much more reason to report. I think it's going to work. And that has some PMFI think obviously has PMF like I, I would say. It's I, I went to this exercise of trying to handicap how much money open AI made from launching open ai deep research.

[00:25:25] I think it's billions. Like the, the, the mo the the she upgrade from like $20 to 200. It has to be billions in the R off. Maybe not all them will stick around, but like that is some amount of PMF that is didn't they have to immediately drop it down

[00:25:38] Jacob: to the $20 tier?

[00:25:39] swyx: They expanded access. I don't, I wouldn't say, which I thought was

[00:25:42] Jacob: really telling of the market.

[00:25:43] Right. It's like where you have a you know, I think it's gonna be so interesting to see what they're actually able to get in that 200 or $2,000 tier, which we all think is, is, you know, has a ton of potential. But I thought it was fascinating. I don't know whether it was just to get more people exposure to it or the fact that like Google had a similar product obviously, and, and other folks did too.

[00:25:59] But [00:26:00] it was really interesting how quickly they dropped it down.

[00:26:02] swyx: I don't, I think that's just a more general policy of no matter what they have at the top tier, they always want to have smaller versions of that in the, in the lower tiers. Yeah. And just get people exposure to it. Just, yeah, just get exposure.

[00:26:12] The brand of being first to market and, and like the default choice Yeah. Is paramount to open ai

[00:26:18] Jacob: though. I thought that whole thing was fascinating 'cause Google had the first product, right? Yeah. And no, like, you know, I, we

[00:26:24] swyx: interviewed them. I, I, I, straight up to their faces, I was like, opening, I mocked you.

[00:26:28] And they were like, yeah, well, actually curious, what's

[00:26:30] Jacob: it, this is totally off topic, but whatever. Like, what is it going to take for go? Google just released some great models like a, a few weeks ago. Like I feel like it's happening. The stuff they're shipping is really cool. It's happening. Yeah, but I, I, I also, I feel like at least in the, you know, broader discourse, it's still like a drop in the bucket relative to

[00:26:45] swyx: Yeah.

[00:26:45] I mean, I, I can riff on, on this. I, I, but I, I think it's happening. I think it takes some time, but I am, like my Gemini usage is up. Like, I, I use, I use it a lot more for anything from like summarizing YouTube videos to the [00:27:00] native image generation Yeah. That they just launched to like flash thinking.

[00:27:02] So yeah, multi-mobile stuff's great. Yeah. I run you know, and I run like a daily sort of news recap called AI news that is, 99% generated by models, and I do a bake off between all the frontier models every day. And it's every day. Like does it switch? I manual? Yes, it does switch. And I, man, I manually do it.

[00:27:18] And flash is, flash wins most days. So, so like, I think it's happening. I think I was thinking, I was thinking about tracking myself like number of opens of tragedy, g Bt versus Gemini. And at some point it will cross. I think that Gemini will be my main and, and it, it, I I like that will slowly happen for a bunch of people.

[00:27:37] And, and, and then that will, that'll shift. I, I think that's, that's a really interesting for developers, this is a different question. Yeah. It's Google getting over itself of having Google Cloud versus Vertex versus AI studio, all these like five different brands, slowly consolidating it. It'll happen just slowly, I guess.

[00:27:53] Alessio: Yeah.

[00:27:54] Yeah. I, I mean, another good example is like you cannot use the thinking models in cursor. Yeah. And I know [00:28:00] Logan killed Patrick's that they're working on it, but I, I think there's all these small things where like if I cannot easily use it, I'm really not gonna go out of my way to do it. But I do agree that when you do use them, their models are, are great.

[00:28:12] So yeah. They just need better, better bridges.

[00:28:15] swyx: You had one of the questions in the prep.

[00:28:16] Debating Public Companies: Google vs. Apple

[00:28:16] swyx: What public company are you long and short and minus Google versus, versus Apple, like, long, short. That was also my

[00:28:23] Jacob: combo. I, I feel like, yeah, I mean, it does feel like Google's really cooking right now.

[00:28:26] swyx: Yeah. So okay, coming back to what has product market fit

[00:28:29] Jacob: now,

[00:28:29] swyx: now that we come

[00:28:30] Jacob: back to my complete total sidetrack,

[00:28:33] Customer Support and AI's Role

[00:28:33] swyx: there's also customer support.

[00:28:35] We were talking on, on the car about Decagon and Sierra, obviously Brett, Brett Taylor is founder of Sierra. And yeah, it seems like there's just this, these layers of agents that'll like, I think you just look at like the income statement or like the, the org chart of any large scaled company and you start picking them off one by one.

[00:28:51] What like is interesting knowledge work? And they would just kind of eat. Things slowly from the outside in. Yeah, that makes sense.

[00:28:57] Alessio: I, I mean, the episode with the, [00:29:00] with Brett, he's so passionate about developer tools and Yeah. He did not do a developer tools. We spent like two hours talking about developer tools and like, all, all of that stuff.

[00:29:10] And it's like, I, they a customer support company, I'm like, man, that says something. You know what I mean? Yeah. It's like when you have somebody like him who can like, raise any amount of money from anybody to do anything. Yeah. To pick customer support as the market to go after while also being the chairman of OpenAI, like that shows you that like, these things have moats and have longstanding, like they're gonna stick around, you know?

[00:29:32] Otherwise he's smarter than that. So yeah, that's a, that's a space where maybe initially, you know, I would've said, I don't know, it's like the most exciting thing to, to jump into, but then if you really look at the shape of like, how the workforce are structured and like how the cost centers of like the business really end up, especially for more consumer facing businesses, like a lot of it goes into customer support.

[00:29:54] AI's Impact on Business Growth

[00:29:54] Alessio: All the AI story of the last two years has been cost cutting. Yeah. I think now we're gonna switch more towards growth revenue. [00:30:00] Totally. You know, like you've seen Jensen, like last year, GTC was saying the more you buy, the more you save this year is that the more you buy, the more you make. So we're hot off the

[00:30:08] Jacob: press.

[00:30:10] We were there. We were there. Yeah. I do think that's one of the most interesting things about the, this first wave of apps where it's like almost the easiest thing that you could you could get real traction with was stuff that, you know, for lack of a better way to frame it, like so that people had already been comfortable outsourcing the BPOs or something and kind of implicitly said like, Hey, this is a cost center.

[00:30:24] Like we are willing to take some performance cut for cost in the past. You know, the, the irony of that, or what I'm really curious to see how it plays out is, you know, you, you could imagine that is the area where price competition is going to be most fierce because it's already stuff that you know, that people have said, Hey, we don't need the like a hundred percent best version of that.

[00:30:42] And I wonder, you know, this next wave of apps. May prove actually even more defensible as you get these capabilities that actually are, you know, increased top line or whatnot where you're like, you take ai, go to market, for example. Like you're, you'd pay like twice as much for something that brought, like, 'cause there's just a kind of very clean ROI story to it.

[00:30:59] And so [00:31:00] I wonder ultimately whether the, like this next set of apps actually ends up being more interesting than the, than the first wave.

[00:31:05] Alessio: Yeah,

[00:31:05] Voice AI and Scheduling Solutions

[00:31:05] Jordan: I think a lot of the voice AI ones are interesting too, because you don't need a hundred percent precision recall to actually, you know, have a great product.

[00:31:12] And so for example, we looked into a bunch of you know, scheduling intake companies, for example, like home services, right? For electricians and stuff like that. Today they miss 50% of their calls. So even if the AI is only effective, say 75% of the time, yeah, it's crazy, right? So if it's effective 75% of the time, that's totally fine because that's still a ton of increased revenue for the customer, right?

[00:31:32] And so you don't need that a hundred percent accuracy. Yeah. And so as the models. And the reliability of these agents are getting better is totally fine, because you're still getting a ton of value in the meantime.

[00:31:41] swyx: Yeah. One, this is, I don't know how related this is, but I, one of my favorite meetings at it is related one of my favorite meetings at AI Engineer Summit, it is like, like I do these, this is our first one in New York, and I it is like met the different crew than, than you meet here.

[00:31:55] Like everyone here is loves developer tools, loves infra over there. They're actually more interested in [00:32:00] applications. It's kind of cool. I met this like bootstrap team that, like, they're only doing appointment scheduling for vets. They, they, yeah. And like, they're like, this is a, this is an anomaly. We don't usually come to engineering summits 'cause we usually go to vet summits and like talk to the, they're, they're like, you know, they, they're, they're literally, I'm sure it's a

[00:32:16] Jordan: massive pain point.

[00:32:17] They're willing to pay a lot of money.

[00:32:20] Alessio: Yeah. But, but, but this is like my point about saving versus making more, it's like if an electrician takes two x more calls, do they have the bandwidth? To actually do two X more in-house and they get higher. Well, yeah, exactly. That's the thing is like, I don't think today most businesses are like structured to just like overnight two, three x the band, you know?

[00:32:38] I think that's like a startup thing. Like mo most businesses then you make an

[00:32:42] swyx: electrician agent. Well, no, totally. That's how do you, how do you recruiting agent for electrician, for like

[00:32:49] Alessio: electrician. Great. That's a good point. How do you do lambda school for electrician? I, it's hilarious.

[00:32:53] Jacob: Whack-a-mole for the bottlenecks in these businesses.

[00:32:55] Like as, oh, now we have a ton of demand. Like, cool. Like where do we go?

[00:32:58] swyx: Yeah.

[00:32:59] Exploring AI Applications in Various Fields

[00:32:59] swyx: So just to [00:33:00] round out the, the this PMF thing I think this is relevant in a certain sense of, like, it's pretty obvious that the killer agents are coding agents, support agents, deep research, right? Roughly, right. We've covered all those three already.

[00:33:10] Then, then, then you have to sort of be, turn to offense and go like, okay, what's next? And like, what, what about, I

[00:33:16] Jacob: mean, I also just like summarization of, of voice and conversation, right? Yep. Absolutely. We actually had that on there. I

[00:33:21] swyx: just, I didn't put it as agent. Because seems less agentic, you know? But yes, still, still a good AI use case.

[00:33:26] That one I, I've seen I would mention granola and what's the other one? Monterey, I think a bridge was one wanted to mention. I was say bridge. Yeah, bridge. Okay. So I'll just, I'll call out what I had on my slides. Yeah. For, for the agent engineering thing. So it was screen sharing, which I think is actually kind of, kind of underrated.

[00:33:42] Like people, like an AI watching you as you do your work and just like offering assistance outbound sales. So instead of support, just being more outbound hiring, you say

[00:33:51] Jacob: outbound sales has brought a market fit?

[00:33:53] swyx: No, it, it, it will, it's come out. Oh, on the comp. Yeah. I was totally agree with that. Yeah. Hiring like the recruiting side education, like the, [00:34:00] the sort of like personalized teaching, I think.

[00:34:02] I'm kind of shocked we haven't seen more there. Yeah. Yeah. I don't know if that's like, like it's like Duolingo is the thing. Amigo.

[00:34:08] Jacob: Yeah. I mean, speak in some of these like, you know,

[00:34:10] swyx: speak, practice, yeah. Interesting. And then finance, I, there's, there's a ton of finance cases that we can talk about that and then personal ai, which we also had a little bit of that, but I think personal AI is a harder to monetize, but I, I think those would be like, what I would say is up and coming in terms of like, that's what I'm currently focusing on.

[00:34:27] Jacob: I feel like this question's been asked a few different ways but I'm, I'm curious what you guys think it's like, is it like, if we just froze model capabilities today, like is there, you know, trillions of dollars of application value to be unlocked? Like, like AI education? Like if we just stopped today all model development, like with this current generation of models, we could probably build some pretty amazing education apps.

[00:34:44] Or like, how much of this, how much of, of all this is like contingent upon just like, okay, people have had two years with GBT four and like, you know, I don't know, six months with the reasoning models, like how much is contingent upon it just being more time with these things versus like the models actually have to get better?

[00:34:58] I dunno, it's a hard question, so I'm gonna just throw it [00:35:00] to you.

[00:35:00] Alessio: Yeah. Well I think the societal thing, it's maybe harder, especially in education. You know, like, can you basically like Doge. The education system. Probably you should, but like, can you, I I think it's more of a human,

[00:35:14] Jacob: but people pay for all sorts of like, get ahead things outside of class and you know, certainly in other countries there's a ton of consumer spend and education.

[00:35:21] It feels like the market opportunity is there.

[00:35:23] swyx: Yeah. And, and private education, I think yeah, public Public is a very different, yeah. One of my most interesting quests from last year was kind of reforming Singapore's education system to be more sort of AI native, just what you were doing on the side while you were Yes.

[00:35:38] That's a great, that's a great side quest. My stated goal is for Singapore to be the first country that has Python as a first language, as a, as a national language. Anyway, so, but the, the, the, the defense, the pushback I got from Ministry of Education was that the teachers would be unprepared to do it.

[00:35:53] So it's like, it was like the def the, like, the it was really interesting, like immediate pushback. Was that the defacto teachers union being like, [00:36:00] resistant to change and like, okay. It's that that's par for the course. Anyway, so not, not to, not to dwell too much on that, but like yeah, I mean, like, I, I think like education is one of those things that pe everyone, like has strong opinions on.

[00:36:11] 'cause they all have kids, all be the education system. But like, I think it's gonna be like the, the domain specific, like, like speak like such a amazing example of like top down. Like, we will go through the idea maze and we'll go to Korea and teach them English. Like, it's like, what the hell? And I would love to see more examples of that.

[00:36:29] Like, just like really focus, like no one tried to solve everything. Just, just do your thing really, really well

[00:36:34] Defensibility in AI Applications

[00:36:34] Jacob: on this trend of of, of difficult questions that come up. I'm gonna just ask you the one that my partners like to ask me every single Monday, which is how do you think about defensibility at the at the app layer?

[00:36:41] Alessio: Oh

[00:36:41] Jacob: yeah, that's great. Just gimme an answer. I can copy paste and just like, you know, have network effects. Auto, auto response.

[00:36:47] swyx: Honestly like network effects. I think people don't prioritize those enough because they're trying to make the single player experience good. But then, then they neglect the [00:37:00] multiplayer experience.

[00:37:00] I think one of the I always think about like load-bearing episodes, like, you know, as, as park that you do one a week and like, you know, some of those you don't really talk about ever again. And others you keep mentioning every single podcast. And one of the, this is obviously gonna be the last one. I think the recap episodes for us are pretty load-bearing.

[00:37:15] Like we, we refer to them every three months or so. And like one of them I think for us is Chai for me is chai research, even though that wasn't like a super popular one among the broader community outside of Chai, the chai community, for those who don't know, chai Research is basically a character AI competitor.

[00:37:32] Right. They were bootstraps, they were founded at the same time and they have out outlasted character of de facto. Right. It's funny, like I, I would love to ask Mil a bit more about like the whole character thing, but good luck getting past the Google copy. But like, so he, like, he, like he doesn't have his own models, basically he has his own network of people submitting models to be run.

[00:37:54] And I think like. That is like short term going to be hurting him because he doesn't have [00:38:00] proprietary ip. But long term he has the network network effect to make him robust to any changes in the future. And I think, like I wanna see more of that where like he's basically looking himself as kind of a marketplace and he's identified the choke point, which is will be app or the, the sort of protocol layer that interfaces between the users and the model providers.

[00:38:18] And then make sure that the money kind of flows through and that works. I, I wish that more AI builders or AI founders emphasize network effects. 'cause that that's the only thing that you're gonna have with the end of the day. Yeah. And like brand deeds into network effects you.

[00:38:34] Jacob: Yeah, I guess you know, harder in, in the enterprise context.

[00:38:36] Right. But I mean, I feel, it's funny, we do this exercise and I feel like we talk a lot about like, you know, obviously there's, you know kind of the velocity and the breadth you're able to kind of build of product surface area. There's just like the ability to become a brand in a space. Like, I'm shocked that even in like six, nine months, how an individual company can become synonymous with like an entire category.

[00:38:52] And like, then they're in every room for customers and like all the other startups are like clawing their way to try and get in like one, you know, 20th of those rooms.

[00:38:59] Jordan: There's a [00:39:00] bunch of categories where we talk about an IC and it's like, oh, pricing compression's gonna happen, not as defensible. And so ACVs are gonna go down over time.

[00:39:08] In actuality, some of these, the ACVs have doubled, we've seen, and the reason for that is just, you know, people go to them and pay for that premium of being that brand.

[00:39:16] Jacob: Yeah. I mean, one thing I'm struck by is there's been, there was such a head fake in the early days of, of AI apps where people were like, we want this amazing defensibility story, and then what's the easiest defensibility story?

[00:39:24] It's like, oh, like. Totally unique data set or like train your own model or something. And I feel like that was just like a total head fake where I don't think that's actually useful at all. It's the much less, you sound much less articulate when you're like, well the defensibility here is like the thousand small things that this company does to make like the user experience design everything just like delightful and just like the speed at which they move to kind of both create a really broad product, but then also every three, six months when a new model comes out, it's kind of an existential event for like any company.

[00:39:49] 'cause if you're not the first to like figure out how to use it, someone else will. Yeah. And so velocity really matters there. And it's funny in in, in kinda our internal discussions, we've been like, man, that sounds pretty similar to like how we thought about like application SaaS [00:40:00] companies. That there isn't some like revolutionary reason you don't sound like a genius when you're like, here's applications why application SaaS company A is so much better than B.

[00:40:07] But it's like a lot of little things that compound over time.

[00:40:10] Infrastructure and AI: Current Trends

[00:40:10] Jacob: What about the infrastructure space, guys? Like I'm curious you know. What, how do you guys think about where the interesting categories are here today and you know, like where, where, where do you wanna see more startups or, or where do you think there are too many?

[00:40:21] Alessio: Yeah. Yeah, we call it kind of the L-L-M-O-S. But I would say

[00:40:24] swyx: not we, I mean Andre, Andre calls it LMOS

[00:40:27] Alessio: Well, but yeah, we, well everyone else just copies whatever two. And Andre, the three of you call it the LMO. Well, we have just like four words of ai framework Yeah. Yeah. That we use. And LM Os is one of them, but yeah, I mean, code execution is one.

[00:40:39] We've been banging the drum, everybody now knows where investors in E two B. Mm-hmm. Memory, you know, is one that we kind of touched on before. Super interesting search we talked about. I, I think those are more not traditional infra, not like the bare metal infra. It's more like the infra around the tools for agents model, you know?

[00:40:57] Which I think is where a lot of the value is gonna [00:41:00] be. The security

[00:41:00] swyx: ones. Yeah.

[00:41:01] Alessio: Yeah. And cyber security. I mean there's so much to be done there. And it's more like basically any area where. AI is being used by the offense. AI needs to be applied on the defense side, like email security, you know, identity, like all these different things.

[00:41:16] So we've been doing a lot there as well as, you know, how do you rethink things that used to be costly, like red teaming and maybe used to be a checkbox in the past Today they can be actually helpful. Yeah. To make you secure your app. And there's this whole idea of like, semantics, right? That not the models can be good at.

[00:41:32] You know, in the past everything is about syntax. It's kind of like very basic, you know, constraint rules. I think now you can start to infer semantics from things that are beyond just like simple recognition to like understanding why certain things are happening a certain way. So in the security space, we're seeing that with binary inspection, for example.

[00:41:51] Like there's kinda like the syntax, but then there are like semantics of like understanding what is the scope overall really trying to do. Even though this [00:42:00] individual syntax, it's like seeing something specific. Not to get too technical, but yeah, I, I think infra overall, it's like a super interesting place if you're making use of the model, if you're just, I'm less bullish.

[00:42:13] Not, not that it's not a great business, but I think it's a very capital intensive business, which is like serving the models. Mm-hmm. Yeah. I think that infra is like, great people will make money, but yeah. I, I, I don't think there's as much of a interest from, from us at

[00:42:25] Jordan: least. Yeah. How, how do you guys think about what OpenAI and the big research labs will encompass as part of the developer and infra category?

[00:42:31] Yeah.

[00:42:31] Alessio: That, that's why I, I would say I search is the first example of one of the things we used to mention on, you know, we had X on the podcast and perplexity obviously as a, as an API. The basic idea

[00:42:44] swyx: is if you go into like the chat GBT custom GPT builder, like what are the check boxes? Each of them is a startup.

[00:42:50] Alessio: Yeah. And, and now they're also APIs. So now search is also an a p, we will see what the adoption is. There's the, you know, in traditional infra, like everybody wants to be [00:43:00] multi-cloud, so maybe we'll see the same Where change GPD search or open AI search. API is like, great with the open AI models because you get it all bundled in, but their price is very high.

[00:43:11] If you compare it to like, you know, XI think is like five times the, the price for the same amount of research, which makes sense if you have a big open AI contract. But maybe if you're just like pick and best in breed, you wanna compare different ones. Yeah. Yeah, they don't have a code execution one.

[00:43:26] I'm sure they'll release one soon. So they wanna own that too, but yeah. Same question we were talking about before, right? Did they wanna be an API company or a product company? Do you make more money building Tri g BT search or selling search? API?

[00:43:38] swyx: Yeah. The, the broader lesson, instead of like going, we did applications just now.

[00:43:42] And then what do you think is interesting infrastructure? Like it's not 50 50, it's not like equal weighted, like it, it's just very clearly the application layer has like. Been way more interesting. Like yes, there, there's interesting in infrastructure plays and I even want to like push back on like the, the, the whole GPU serving thing because like together [00:44:00] AI is doing well, fireworks, I mean I was, that worked.

[00:44:02] Alessio: It's like data

[00:44:02] Jacob: centers

[00:44:03] Alessio: and inference

[00:44:03] Jacob: providers,

[00:44:04] Alessio: the,

[00:44:04] swyx: you know,

[00:44:04] Alessio: I think it's not like the capital

[00:44:06] swyx: Oh, I see.

[00:44:07] Alessio: I for, for again, capital efficiency. Yeah. Much larger funds. So you, I'm sure you have GPU clouds. Yeah.

[00:44:13] swyx: Yeah. So that's, that's, that is one thing I have been learning in, in that you know, I think I have historically had dev tools and infra bias and so has he, and we've had to learn that applications actually are very interesting and also maybe kind of the killer application of models in a sense that you can charge for utility and not for cost.

[00:44:33] Right? Which, where like most infrastructure reduces to cost plus. Yeah. Right. So, and like, that's not where you wanna be for ai. So that's, that's interesting for, for me I thought it would be interesting for me to be the only non VC in the room to be saying what is not investible. 'cause like then I then, you know, you can I, I won't be canceled for saying like, your, your whole category is, we have a great thing where like, this thing's

[00:44:54] Jacob: not investible and then like three months later we're desperately chasing.

[00:44:56] Exactly. Exactly. So you don't wanna be on a record space changes so [00:45:00] fast. It's like you gotta, every opinion you hold, you have to like, hold it quite loosely. Yeah.

[00:45:02] swyx: I'm happy to be wrong in public, you know, I think that's how you learn the most, right? Yeah. So like, fine tuning companys is something I struggled with and still, like, I don't see how this becomes a big thing.

[00:45:12] Like you kind of have to wrap it up in a broader, ser broader enterprise AI company, like services company, like a writer, AI where like they will find you and it's part of the overall offering. Mm-hmm. But like, that's not where you spike. Yeah, it's kind of interesting. And then I, I'll, I'll just kind of AI DevOps and like, there's a lot of AI SRE out there seems like.

[00:45:32] There's a lot of data out there that that should be able to be plugged into your code base or, or, or your app to it's self-heal or whatever. It's just, I don't know if that's like, been a thing yet. And you guys can correct me if you're, if I'm wrong. And then the, the last thing I'll mention is voice realtime infra again, like very interesting, very, very hot.

[00:45:49] But again, how big is it? Those are the, the main three that I'm thinking about for things I'm struggling with.

[00:45:54] Jordan: Yeah. I guess a couple comments on the A-I-S-R-E side. I actually disagree with that one. Yeah. I think that the [00:46:00] reason they haven't sort of taken off yet is because the tech is just not there quite yet.

[00:46:04] And so it goes back to the earlier question, do we think about investing towards where the companies will be when the models improve versus now? I think that's going to be, in short term we'll get there, but it's just not there just yet. But I think it's an interesting opportunity overall.

[00:46:18] swyx: Yeah. It's my pushback to you is, well it's monitoring a lot of logs, right?

[00:46:22] Yeah. And it's basically anomaly detection rather than. Like there's, there's a whole bunch of like stuff that can happen after you detect the anomaly, but it's really just an anomaly detection. And we've always had that, you know, like it's, this is like not a Transformers LLM use case. This is just regular anomaly detection.

[00:46:38] Jordan: It's more in terms of like, it's not going to be an autonomous SRE for a while. Yeah. And so the question is how, how much can the latest sort of AI advancements increase the efficacy of going, bringing your MTTR down? Yeah. And I see even if it's 10% improvement on beforehand, it's still potentially a lot of revenue.

[00:46:55] swyx: Okay.

[00:46:56] Jordan: Right. That's the way, at least I, I think I would think about it now and then, you [00:47:00] know, a few years from now, if it's actually an autonomous SRE just replacing altogether, then that's a totally different thing.

[00:47:05] swyx: Hmm. Cool. I, I look after it.

[00:47:08] Jacob: Yeah. Yeah.

[00:47:08] Challenges and Future of AI

[00:47:08] Jacob: You know, I guess switching back to overly broad questions, like what do you feel like is the biggest unanswered question in AI today, or, you know, that has, and, you know, large implications for the ecosystem?

[00:47:17] I.

[00:47:17] Alessio: Yeah, I, I've been banging the drum on RL and I think it's clear that you can do RL successfully on verifiable domains. Yeah. I would say whether or not we can figure out how to do that in non-verifiable one. So law is a great example. Totally. Like can you do RL on contracts and documents? Marketing sales, going back to outbound sales, like can you do RL to like simulate what an outbound and, and kind of like the conversation leads to yeah, it's unclear.

[00:47:45] If not, then I think we'll be stuck with like, you're gonna have agents in the more verifiable domains and then you'll just kinda have copilots. And the non-verifiable ones because still you'll still need a person to be the tastemaker.

[00:47:56] Jacob: So I had the exact same thing and I feel it's like the que, I just, I'm trying to think of the [00:48:00] implications where if it doesn't work, like the world could be weird, where like you have like fully autonomous AI coders and like, you know, no one does any software or math or even like, you know, some areas of science, but then like to write the most basic sales email still like, like just, it's always so hard to predict how the world like that is such a weird of all the sci-fi that was written, you know, 50 years ago, I don't think anybody foresaw that future.

[00:48:21] That is a really weird future. Yeah,

[00:48:22] swyx: I, I've called it industrialized autism. Do either of you

[00:48:25] Jordan: have a different one for that Think unanswered question, I guess? I dunno if this is a good answer, but you know, Bob McGrew we had on the podcast and he was talking about like the rule of nines they have at open ai where to go from 90% reliability to 99, it's an order of magnitude increase in compute and then 99 to 99.9 order of magnitude increase.

[00:48:41] And that happens every two to three years. And so I think how are we going to scale sort of accordingly? This sort of next part, I think there's a lot of unanswered questions, just like from a hardware perspective. And then I think as part of that, from the availability perspective, like is Nvidia just going to continue to be dominant?

[00:48:59] Like obviously [00:49:00] AWS is going hard into what's their train chips? Mm-hmm. I'm blank on that. Thank you. And so I think like there's a big ecosystem around kuda that's obviously allowed Nvidia to remain dominant, but just what's going to happen and is there anyone's going? Is there anyone that's going to come sort of combat that to increase the availability GPUs?

[00:49:21] Or are we just gonna be constrained going forward when we actually need way more compute going forward?

[00:49:25] swyx: Yeah. My quick thoughts. I've, I've been I'm the only individual named, there's an investor in Medex which is kind of like really funny 'cause everyone else has funds and no knows just me. And and it's, it's, it's, there's a, there's an interest, like there are all these like Nvidia startups, like, sorry dedicated silicon startups that are coming up and, and trying to challenge that.

[00:49:44] And the simple answer is like these GPUs are the most general thing is possible by, by design. That's why they do gaming and crypto and ai. And I think as long as the architecture seems, seems stable, it seems like there's a case to be made for for that. The only question is who will [00:50:00] win that?

[00:50:00] And obviously there's a whole bunch of competitors, including I think AMD's trying to, you know, to, to make a play for it. But so will AWS and so will, you know, every other, like Microsoft has a chip, Facebook has a chip so who knows who will win that. It, it just, it's very interesting that like, this seems to be such a valuable prize.

[00:50:18] Like it's. Freaking Nvidia that you're competing with. Yeah. And no one has really made a real d there yet. But so I, I kind of, I kind of agree with you, but like, I think that basically it's all about stability of workload and as long as it's a bet on like the depth of transformers basically. And if you are fine with that, like even and I think like the, even a state space model, people would agree that like, it, it wouldn't really change that much.

[00:50:41] And probably I think the, the overall consensus is that you don't even use state space models in individually. Like you would use them in a mixture with transformers anyway. Yeah. So then like, yeah, just go bet on transformers and bake it into the chip and you'll have much, way more you know, basically Asics like for transformers and that's fine.

[00:50:59] [00:51:00] And, and so like prima fassy, there should be a company that wins that. Yeah. I don't know who will win. Yeah. I wish we knew. Yeah, I think that, I think anyone, you have to start basically after 20 19, 20 20. Because anyone started before that, we'll still be too general. Mm-hmm. Yeah. Yeah. Because you, transformers hadn't won yet one at the time.

[00:51:19] I have one more. It, the, the, I think that the most emergent one that came out of the New York conference that I did was age agent authentication. Mm. I think literally the, like the information just published a, like, this is something that they're worried about, which is when operator or whoever accesses your website on behalf of you, how does it indicate that it's not you, but it's, it's an agent of you.

[00:51:39] And I think like my general philosophy on agent experience or any of the sort of like reinvention of every part of the stack for agents is the all not, not necessary except for this agent off thing. Like we, we really need to, to be able to like new, new SSO effectively for agents.

[00:51:54] Jacob: Is it gonna be crypto?

[00:51:55] Both crypto people are really amped about the

[00:51:57] swyx: you know, like it really, it's [00:52:00] really frustrating when Sam Altman is right, but like, maybe you have to scan your eyeballs. Like maybe you just have to, you just like. Maybe he saw this like five years ago and he was like, you gotta scan your eyeballs. And like the rest of us are just behind him as usual.

[00:52:14] Jacob: I love it.

[00:52:15] Quick Fire Round: Dream Guests and News Sources

[00:52:15] Jacob: Well, okay, now I'm, I'm move to the quick fire round where, we'll, we'll go around the horn and get a, get quick takes on things. So the first is gonna be dream podcast guests, John Carmack.

[00:52:24] swyx: Yeah. He's six. John is like six steps away from solving a GI apparently. So we just ask him how long p he is.

[00:52:30] Exactly. For us it's Andre. For me it's Andre. He had, he's a listener and supporter of the pod. And like basically when I launched the, the whole AI engineer push that we, we, that we have, he was the basically the first one to legitimize it. He was like, you know, there will be more AI engineers than ML engineers.

[00:52:46] And I think that made everyone else pay attention. So like, lean Space only exists because, you know, he, he helps he and other people help to promote it. Yeah.

[00:52:54] Jacob: I also had Andrea, so Yeah, thinking the same thing there. I basically, mine's a little bit cheating, [00:53:00] but I think at some point there will clearly, like they're writing a book about OpenAI now, and at some point, like somebody probably acquired will get to do the acquired OpenAI episode.

[00:53:06] But if unsupervised learning could, like, there's clearly just like so many amazing stories of like the last five, six years.

[00:53:13] So do you know

[00:53:14] swyx: about Doomers? It's a play, I'm actually going to it this Saturday. Yeah, they, they, they, they, someone made a play about the board drama of last year. Really?

[00:53:22] Of two years ago. Wow. Yeah. Wow. That's cool. That's not how it's, yeah, let us know. There will be, we should director of that on the podcast. I, I think it's a lot of fan fiction, basically, but like, someone will write the accounts and, and it'll be interesting and fascinating and a lot of, a lot of it will be fake because it's a complex beast, right.

[00:53:40] You're just getting an oral history of what happened.

[00:53:42] Jacob: Yeah, yeah. Yeah. Alright. For the next one, I figured you could shout out either like a new source you used to stay up to date or a startup that's, that you're not invested in, that you're excited about. Oh. Or you can

[00:53:52] Alessio: do it. My news sources is Sean, that's what I was gonna say.

[00:53:56] I literally wrote swix Twitter. So in our, in our disco, we [00:54:00] have a space, wait, we have a latent space discord. Any link that ever matters on the internet, Sean's gonna post it in the Discord. So all I do, I open Discord and, and we have like, you know, 40, 50 different channels by topic. Actually that is very true.

[00:54:14] That's very true. I opened Discord and I'm like, okay, ai. Then I go developer tools, then I go Creator Economy, then I go stock and macro. Then I go and they're all there. So thank you. Yeah,

[00:54:24] swyx: we actually met because of the Discord. It was like a Covid thing 'cause everyone's at home. They just started a discord and yeah, that was the origin of Living Space.

[00:54:32] Just chatting on the Discord. It

[00:54:33] Alessio: used to be called Dev slash Invest. Yeah. So it was all about developer tools investing. And then we were open AI in October, 2022. We were like, maybe we should do a podcast. And that open AI was the first guest history. Yeah.

[00:54:44] Yeah. Yeah.

[00:54:45] swyx: Yeah. I I was not prepared about the, the news sources thing.

[00:54:50] I think maybe it's hard, it's really shitty to say, but like, just in person conversations. Yeah. And I think the reason I have to be [00:55:00] here in SF is because I make friends with people who know things and are smarter than me, and we do go for chats and they're nice enough to share some stuff. And so sometimes I wish I, I worry that I am being used in order to put things out there that are maybe not true, but, you know, so I have to exercise my own judgment as to what that is.

[00:55:21] I think one of the cool

[00:55:21] Jacob: things about the podcast in general is just like the opportunity to take these conversations that happen in like closed rooms and, and try and bring them on to, to the airways. I'm curious like how much of what you, how, how much do you feel like the private discourse is similar to the, to the public discourse?

[00:55:35] swyx: In, in, in many ways the, it is surprisingly similar.

[00:55:39] misc: Yeah.

[00:55:39] swyx: As in. People at OpenAI learn about things about OpenAI from us, which is interesting. And then there are some ways which is drastically not drastically dissimilar. And those are the things I just cannot repeat until it's public.

[00:55:52] Final Thoughts and Plugs

[00:55:52] Jacob: This has been super fun.

[00:55:53] I thought I lived up to it. We were looking forward to this for a while. We wanna make sure everyone around the horn gets an opportunity to plug whatever they wanna [00:56:00] plug. So we'll leave the last word to to all of us, I guess. Where can folks go to learn more about latent space and all the exciting things you do?

[00:56:06] Wanna make sure our listeners have a good sense of everything?

[00:56:08] Alessio: Yes. So we have a substack latent. That space is the website. And then please subscribe on YouTube. We're doing a lot of YouTube. We're trying to do better video and all that. So he set our OKRs and,

[00:56:20] swyx: It's, it's basically all YouTube.

[00:56:21] Alessio: Come, come watch us on YouTube. It's very important for me personally, even if you don't care. Just OKRs. Well, we have to

[00:56:28] swyx: increase our production value. Look at this. I know, I know. We only have three cameras.

[00:56:34] Alessio: Yeah. And then Sean does a lot of the writing outside of the podcast on the newsletter, so

[00:56:38] swyx: yeah, so it is like trying to be newsletter and community and podcast and whatever else that we do.

[00:56:46] Yeah, so I guess for, for me, I guess there's the in space, but then there's also the other big piece, which is the, the conference that I run. Yeah. And the idea is that I think sometimes you just get the, the good stuff from people if you just put them in front of a lot of [00:57:00] people. And that's really like, I'm mining people for content and sometimes you put a mic in front of them and they yap for an hour.

[00:57:05] Other, other times you have to put them in front of like a prestigious conference and then they drop some alpha. And so the next one for us is gonna be June. It's the AI Engineering World's Fair. And it should be the largest technical conference

[00:57:18] Jacob: for ai. And ours is simple. Just we, we were, we just run a humble podcast.

[00:57:22] So no subscribe to unsupervised learning on YouTube. Fixed. Thanks so much. This was this was awesome. Thanks for having me. It was good to see you guys. Thanks for coming on.