The Exchanges

Q And the, the, the real kind of question then is to match all of that capital, to match all of that need, is the demand actually there?

A Yes. Yes. That's become, uh, the six hundred billion dollar Question. So named after, uh, you know, a great essay by David Kahn. Uh, but you know, not just David, there's a Goldman Sachs had a report a few months ago, sort of, uh, questioning, uh, the potential imbalance between demand for AI and, uh, the infrastructure that is being built. So, you know, as per the above, like people are building and building and building. The question is indeed, uh, whether, uh, you know, uh, people will, uh, have, uh, a need for AI that will match, uh, what, um, what is being, uh, Built. And, uh, what was particularly hard to nail is the timing of those things, right? Because to build a supercomputer, to build a data center, to build chips, like all of this are, you know, two, three, four year commitments. This is a lot of hardware that you need to like build and put together and, you know, order the parts and like all the things. And, uh, that's a particularly sort of unforgiving kind of a kind of cycle. So if you miss time, uh, demand and, Even if you get demand, the right amount of demand, eventually there's a gap between when that demand actually materializes and all the money that you spend up front, then you could find yourself in potential trouble, at least short term.

AI assessment note: “Yes. Yes. That's become, uh, the six hundred billion dollar Question.”

Q And so far, I think most people have separated AI as a different sort of concept than SAS. What does AI mean for SAS? Is it the death of SAS? Is it an extension?

A Yeah, there is that, that, that whole, uh, that whole discussion. Um, you know, look, we, we, we think the answer is, uh, is no. We think that SAS It's not going to be dead. We think that SAS is just going to have to evolve dramatically. Uh, and if you think of SAS being largely, uh, a, a wrapper around a database, and you could argue that's what Salesforce is, you know, a lot of, like, workflow and on top of, um, on top of a database, the next generation of SAS is going to be a wrapper around intelligence. And, uh, you know, with, uh, workflow, with, uh, integrations and, and, and all the things. Look, that evolution is going to happen. Some existing vendors are going to do it. Some new vendors are going to do it. But we think that, you know, intelligence is going to be everywhere, which is kind of funny, by the way, because we effectively saying that the future is rappers. And, you know, in the last year, every VC would say, well, I don't want to invest in rappers because where's the defensibility and all the thing. And I think the collective thinking has shifted pretty dramatically. And now we've seen a bunch of rappers, you know, specialized For this industry or this problem actually grow very fast and everybody wants to invest in them.

AI assessment note: “We think that SAS It's not going to be dead.”

Q to have a conversation with a potential customer, and you have half an hour. There is no way you can look through all that data to extract what is the essence of that, right? And we help you to do that, right? So it just turned, turn this completely around, and people just love it because Because what, what, how it enables you as a seller to be super effective.

A It's amazing, by the way, to the, you know, ongoing debate that we all have about, um, what happens between AI and, and our jobs, uh, you know, the big thing that you hear a lot, um, is that, um, it will make, uh, uh, okay to mediocre people much, much better, uh, and that, uh, it just like levels up, uh, the state of play across professions, but like in the case of, Of, of, of sales. What you just described seems like a perfect example of this, where you can not be that good at your job, but AI will just basically put you at the level of, like, people who are pretty excellent at the job, at least in terms of, like, preparation and identification of, of targets.

AI assessment note: “What you just described seems like a perfect example of this”

Q Starting with the, the landscape itself, um, you know, why is it so crowded? You know, those logos are still sort of floating around in my head as I try and fall asleep. So, um, why do we have so many companies on it?

A Yeah. So of course that's sort of the, the knee jerk reaction that, uh, anybody that gets exposed to that mad landscape, uh, has for, you know, very obvious reasons for anyone that, uh, has seen the PDF. It is a little bit of an eye chart and every year it keeps getting, uh, worse in, in many ways. You know, I, I'd argue that, um, look, as much as we try to do this as, A map. Uh, it's also a reflection of the territory. In other words, the reason why it's so crowded is not because of us. It's not because we put, uh, you know, gradually all sorts of, uh, logos on, on a map. It's just the, the reality of the industry, uh, that there has been an explosive amount of companies created and founded over the last few years. It's a combination of, of two trends, and maybe we can talk about those in greater detail, but there is one trend that, uh, largely has been You know, played out, I would say between 2015 and 2021, that, uh, we can call the modern data stack trend, which is, uh, you know, all the excitement around, uh, the rise of cloud data warehouses and all the tooling that, uh, went around them, you know, before them, like the ELT, ETL, and after them, uh, reverse ETL, like all the, all the things. And the peak of the excitement was a snowflake IPO. And there was just a tremendous number of companies created tremendous number of companies funded. And so that's one world. And the…

AI assessment note: “the reason why it's so crowded is not because of us. It's just the”

Q So, so seeing some slight cracks, uh, was twenty-twenty-three a head fake? What's going to happen in twenty-twenty-four? Any thoughts around that?

A Yeah, so, look, in, in terms of a head fake, we don't know, uh, what we do know working with, uh, a number of, uh, players Uh, in the enterprise arena in particular is that a lot of, uh, the money that came out, um, that the, that the generating AI purchases came out of were, were innovation budgets rather than actual operational, uh, budgets. And that's, you know, a, a feature, not a bug because that's what, uh, innovation budgets are, are for. Um, but the big question is whether now we are going to go from this kind of innovation budget moment, um, you know, bring in Accenture, let's do a POC kind of thing to now, okay, we're serious. We've identified the use cases. Uh, we grabbing money from, uh, operational budgets where we may actually be removing money from one bucket to put it in this bucket kind of thing. And that's going to be the big, Opportunity and challenge of twenty-twenty-four. So it seems to be moving in that direction, but we, we don't know, we don't know yet.

AI assessment note: “that's going to be the big, Opportunity and challenge of twenty-twenty-four”

Q And, uh, you know, obviously we're, we're VCs, so we, we ask this question every day, but maybe we can talk about it here as well. Um, you know, 20, 23 question mark, 20, 24 question mark. What does that all mean, uh, for us as VCs? How do we think about things?

A You know, we love the space, like, and so the mad, uh, landscape, the mad podcast, like all the, all the work we, we, we, we do in all the investing, um, we do. It's been really interesting To see how, uh, the VC world in general has reacted. There seems to be a whole spectrum. There seems to be, uh, you know, on the one hand, new funds that are entirely specialized, uh, doing basically generative AI. Uh, there seems to be within existing firms, new fund vehicles being raised to do generative AI. That's like the sovereign stuff that, Uh, may or may not be happening like the Saudi Arabia, forty billion AI fund. So there's all of that. And then at the other end of the spectrum, you actually have a handful of, uh, very respected, the venture firms that say, well, we kind of actually are sort of sitting out of the whole thing for now. Uh, and, uh, you know, some, some of them saying, well, our AI strategy is pretty simple. We're just going to dump like a bunch of money into open AI and we're going to make one investment. Uh, which is, uh, which is a, which is a fascinating, uh, concept. You know, where, where we are at first mark as a firm, um, we, uh, I think are, uh, both enthusiastic, but also, uh, careful and ultimately don't feel like we are in, uh, a massive rush. So we sort of keeping our investing pace, uh, consistent with prior years. Um, You know, I think, I think the, th…

AI assessment note: “keeping our investing pace, uh, consistent with prior years.”

Q Absolutely agree. And, uh, you know, we've played around with this topic or maybe talked around it a little bit over the course of, of our chat about AI, but there are other elements of the landscape as well as exciting as AI is. So let's talk modern data stack. What are your thoughts? Is it dead? Is it alive?

A Yeah. And, and, you know, and maybe that's something we, we, um, You could have mentioned at the, at the beginning, but like, look, a big, um, a big reason why, um, we have on the landscape both the, um, All the data infrastructure stuff and all the AI stuff, which is basically the left side of the landscape and then the right side of the landscape is because it's very much a symbiotic relationship between, uh, those two parts and, um, you know, to be, as we were saying, as to be able to feed those models, uh, you need data and data needs to be grabbed from somewhere and needs to be prepped, need to be organized, need to be processed in the, you know, in a certain way. Um, so that's, you know, the, the, the reason why we, On one landscape can have both the generative AI trend and the modern data stack trend. So now to your question about the modern data stack, I don't think that the, the underlying reality of it is, is dead. I think the, you know, the idea that, um, you should have, uh, a bunch of your data in, uh, one of those, uh, amazingly elastic repositories, uh, whether that's, uh, a data warehouse kind of, uh, thing like a snowflake, uh, or redshift, or one of those data lakes, like Like houses, like Databricks, although all of this has been converging. I think that idea is very much as valid as ever and very vibrant. I do think that you need tools that are going to enab…

AI assessment note: “I don't think that the, the underlying reality of it is, is dead.”

Q And so those were some of the arguments for why you might think we are in a bubble. What about the case for not in a bubble?

A We're not in a bubble. Uh, all right. So let's see. I guess, first of all, uh, you could say that many of the AI startups, uh, are doing incredibly well, uh, which seems to, uh, you know, evidence some level of, uh, demand, you know, I would, Even more so, like, rabid interest. Uh, and, um, if you look at OpenAI, for example, a lot of people have, um, argued, including, uh, obviously all the investors, that actually the valuation is not that rich, if you think about it. Uh, so, uh, first of all, the company is growing extraordinarily fast, and the company has extraordinary ambitions to be at a hundred billion dollar in ARR in 29, which is not that far away. Uh, but the multiple Uh, on the, on the deal on a forward revenue basis was 13.5 X. Uh, so, you know, to the discussions that we were having, uh, a minute ago in terms of market multiples, it's actually in the grand scheme of things, not that high. So that's the, that's the argument.

AI assessment note: “the multiple Uh, on the, on the deal on a forward revenue basis was 13.5 X.”

Q Um, so, you know, we talked about kind of OpenAI, one of the major players, but overall, you know, where, where do you think we are in the hype cycle of AI?

A So look, I mean, it feels like, uh, there are some cracks in the, uh, hype cycle. You know, look, there's, there's always a little bit of like, what goes up must come down at some point. And, um, for a society that, uh, as all seems to have a little bit of an ADD problem, feels like we've been all, all of us have been excited about Generative AI for, A number of months now, years or years. So it's gonna be interesting to see like how long that lasts. It seems like it's constantly fueled by the new thing and the new thing and the new thing. And we're like talking about Lama three a second ago, but there's also the, the Microsoft, um, uh, you know, speaking images, uh, thing. And, you know, there's always something new to be excited about, and that has been fueling the hype cycle. At the same time, uh, here and there, this, well, again, what feels like, like, like track. So in particular, there seems to be, uh, perhaps the recognition that if you're not open AI, if you're not anthropic, but you are just a little behind, um, as a startup, things are a bit tougher than they may appear. So I'm, I'm thinking about a couple of recent examples, Inflection AI and stability. Uh, and you know, so Inflection AI, as was abundantly, um, uh, reported in the press, had this, uh, super interesting sort of, uh, acquisition, non-acquisition, which, uh, you know, uh, I tweeted that the, that the M…

AI assessment note: “there are some cracks in the, uh, hype cycle.”

Q There's a lot of overlap too, right? It feels like there's a lot of things converging, a lot of things separating.

A Yes. Yeah, there's, uh, you know, we're, we're like, you and I have, we've talked quite a bit about, um, Uh, the vector databases world. So, vector databases are a key part of the, of the, that emerging modern AI stack because they are the heart of, uh, RAG, retrieval augmented generation, which is basically the way you bring in your data as an enterprise into, uh, a generative AI context, uh, in, in, in particular, not, not just for, but like in particular to avoid hallucinations. So you, you just sort of, um, Uh, you know, compare the, what comes out of the model with like your enterprise data. Um, and, um, you know, vector database is what, what, uh, holds, uh, information or data in a vector format, which is what can be consumed. So, uh, it's sort of the step before you, the LLM in that kind of like modern AI stack architecture. But anyway, the, the, the point being that, uh, there's been like a bunch that, uh, either have a disappeared or have accelerated dramatically in the last couple of years, uh, so the, you know, the pine cones and the quadrants and the chromas and the weviets, um, of, of the world, and I'm sure I'm, I'm forgetting LensDB, and, you know, a lot of very smart entrepreneurs, a lot of, uh, very, uh, great, uh, products, uh, from, from everything we, we're hearing, and a lot of them have had banner years, uh, but equally there's been, like, this emerging q…

AI assessment note: “general purpose players like the MongoDBs of the world have started announcing vector capabilities”

Q Now we can start jumping into the topic du jour, the hot topic of the moment. Um, what's going on with open source AI?

A Yes. So, uh, it's fascinating, right, how we published this one, like two weeks ago, three weeks. So yes, not that long ago. And, um, you know, it's a little bit like the interstellar joke, right? Like when one minute in this world is like, uh, seven years on earth or whatever the exact numbers are. Uh, but it's sort of, Feels like that in, in AI, and, um, certainly last week there was this major announcement by Meta around Llama three, which is basically two and soon three highly performance, apparently, models. It's, uh, seven billion parameters, one, or eight, eight billion, then there's 71, and then there's, uh, four or five billion. Uh, parameters model that, uh, is still being trained. That is like the latest entry into this whole wave of open, uh, source AI, which has been fascinating to watch. And, uh, yeah, that raises a bunch of questions. So this, this one question, and just to stir the pot a little bit, because, uh, you know, we very much love open source in, uh, you know, uh, in everything that we do as, uh, as investors and, and, you know, for, for, for all the, Obvious reasons, or the reasons that have now become obvious, but, uh, having said that, this, this, is that, is that too much of a good thing? There's been like such an explosion of open source in AI that we hearing from customers and users that we talked to that, uh, it's just a little dizzying. Uh, I th…

AI assessment note: “last week there was this major announcement by Meta around Llama three”

Q Yeah, it's yesterday's news. Um, what do we have now? We have artificial super intelligence. So don't care about general human intelligence. We want super intelligence.

A New and improved. Yes. And, uh, yes, sort of the term du jour, and we're talking about Massa, um, you know, a few minutes ago, and that's, that's his thing now. Like he said, like he was sort of born, like his mission in life was to assure the era of, uh, super intelligence. So yeah, that's, that's the term look like. So, you know, again, like, We, we're poking gentle fun at all of this, but, um, you know, we love the space, but, so ASI is the thing, so if you're a new and ambitious company, you need to have superintelligence in your name, so there's Safe Superintelligence, Elias Company, there is another company called Mathematical Superintelligence, like we saw a handful of others, and, you know, it's the same thing as AGI, nobody really has a definition of what artificial superintelligence is, there's like that directionally, that, that, um, General, uh, kind of, like, agreement that super intelligence means intelligence that's better than any human, uh, and, uh, in particular, uh, the ability to, for AI to create more AI and sort of, like, self-perpetuate, or sort of AGI, creating more self-AGI, which would be a characteristic, characteristic of, um, of, uh, of super intelligence.

AI assessment note: “super intelligence means intelligence that's better than any human”

Q you know, the, the kind of highlighting the, the 2023 mantra and mindset, which is AI is going to destroy the world. It's going to kill us all. Um, and now nobody really seems to care, right? You know, AI is doing cool things. People are building on top of it and people are focused on getting that into production versus worrying about whether it's going to kill us all.

A Yeah, it's sort of unclear what happened to doomerism in the last year or so. That was like the dominant discussion not that long ago. I wonder if it's just a function of people just playing around with all those tools, including, you know, ChatGPT and sort of realizing the limitations of it and that, you know, it appears as super intelligent, but it's not. It's, you know, extraordinary in many ways, but it's not a form of a, You know, intelligence. So maybe, maybe it's like that, or, or maybe it's just people actually trying to deploy AI in their enterprise for their specific kind of use case and realizing that, uh, it's actually hard and, uh, actually even deploying AI, uh, there's, there's a lot of things that have nothing to do with AI itself. Like, you know, to get, um, you know, procurement authorization and compliance and all the things. So, uh, you know, far from killing us all, uh, just like deploying it in your department is, uh, is a struggle. So, uh, you know, look, where, where, where does that leave us, uh, in terms of, uh, our own investing? So just sharing, uh, you know, the, the thoughts that it might be interesting for people to hear what, how people like us think about this space, uh, and that's us, but that's a lot of conversations that like VCs have, that founders have, that VCs have with founders. So just like a general sentiment of like what, what, what, …

AI assessment note: “far from killing us all, just like deploying it in your department is a struggle.”

Q Absolutely. And, and maybe to your kind of analogy there between the cloud players and, and others, you know, what's fascinating about, about this industry is how quickly it's been evolving. So think about kind of cloud players took a long time to develop that market share and add those services, but open AI just seems to be moving at a pace. It's insane. What's up with them?

A So that part of the discussion very much falls in the, you know, in, in the, in the camp of like, okay, we're industry observers. Uh, we, we not, Investors in, in OpenAI, uh, but just like everybody else, we've been fascinated with the company, and it's, it's everything, right? It's the scale of ambition, the pace of innovation, and of course, like, all the drama from, uh, you know, the, the, the CEO, Astor, that, uh, that lasted, uh, you know, a minute and a half to safety, uh, developers or engineers that were let go recently. I mean, it's just like, seems like it's Something new happens every minute, and it's fascinating. It truly is amazing to see a company that sort of escapes the, you know, the sphere where we all operate that's dominated by the sort of laws of gravity where, you know, you have to, you're just like raising money is hard and all the things. This is a company that seems to have like unlimited access to like everything, including Capital. And, uh, uh, you know, it's probably a good thing because they presumably are burning immense amount of, of, of money and immense amount of, of, uh, compute, uh, for their very ambitious goals so that it sort of feels like they, they kind of like have to continue, uh, raising, uh, but they're very successful at it. So it's, it's, it's just fascinating to watch. The question is whether they Can continue to escape the laws of…

AI assessment note: “It's the scale of ambition, the pace of innovation, and of course, like, all the drama”

Q Apple Vision Pro, but, um, you know, what, what is, what does that kind of mean for, um, sort of the conversation that we were just having, right? The one around, Having a bunch of different tools across the stack and, uh, full stack AI platforms. Which one do you think is going to win? Is there going to be a winner? Is there going to be room for both?

A Yeah. So, you know, I think we, we certainly are excited about, um, vertical tools or tools that have, um, a pretty kind of narrow scope. You know, there's like this whole discussion around, uh, is, uh, AI going to kill SAS? And, uh, I think the conclusion we came to is that, um, uh, the, the answer is that AI is probably not gonna kill SAS as such, as much as giving rise to a new generation of AI native SAS companies, uh, where, um, I think those companies will have a full stack approach where, um, they, Focus on the application and the workflow and the collaboration as much as the model. Uh, and, uh, so we, we certainly gravitate towards companies where founders can do both, but certainly have the technical chops for going directly into the model and doing, you know, fine tuning rag, customizing with the data, like all sorts of different things one can do at the, at the model level. So I think Those full stack AI native vertical applications are going to be a very exciting area to watch.

AI assessment note: “Those full stack AI native vertical applications are going to be a very exciting area”

Q Our customers frequently tell us that we're a million times faster, right? So when you want to do a recommendation engine, you want to find patterns in, wait, who is Matt similar to, what have they purchased, and how are they connected and connected to the product hierarchy, right? That's typically five, 10, 12, 15 hops in a connected data structure. Like a graph database is freaking amazing at that.

A And just to recap, just to make sure it's clear to everyone. So first of all, graph database, like you coined the term, right, if I remember correctly. And that's when you started the company in like 2007 or something like this. Like you're literally at the origin of the space, which. Which was just your idea and has now become a whole space with like different companies and competitors and all those things where you repioneer all of this, just to put it in context. But, uh, so a graph database is a database that elevates relationships as a first class citizen, as opposed to just like Rose and Collins, like people are, are the, the, the product understands how things entities are connected to one another, right? In the, in the most simple layman's term. Is that, is that correct?

AI assessment note: “a graph database is a database that elevates relationships as a first class citizen”

Q of threaded version of, uh, the thing that happens, which is, did Finn answer the question in such a way that no work was put onto your CS team? If so, then fine. We do report on things like customer satisfaction for Finn users, and our, our customers can see that to make sure that the, that the Finn customers aren't more pissed off than the human ones, et cetera.

A It's a really interesting discussion, like almost like philosophically in terms of like how we measure AI. I think that's a tendency to hold AI to higher standard. Absolutely. But, but, um, you know, in, in this case, uh, from a pricing perspective, you hold AI to the same standard as, as a human. So, you know, one way you could go is like, okay, well you answer the question about how to reset a password, but then track whether the person actually successfully did and that was completely resolved. But, uh, no, I think that's very fair. For what it's worth, uh, that, um, you know, you wouldn't expect AI to do more than, than, than, uh, than a human, uh, from that perspective.

AI assessment note: “you hold AI to the same standard as, as a human”

Q Around AI and things that we're kind of looking at, excited about, you know, there's a stack emerging, right? Just like, you know, maybe there's a parallel to the modern data stack, as we've kind of talked about before, but there's a stack emerging around AI. What does that look like in your view? And, you know, what do you think about that stack in general?

A Yes, we, we used the term, uh, modern AI stack in the write-up, which, uh, had been trying to resist for a while, but like, it's, it's sort of a, it's sort of convenient, and the reason why I'd been trying to avoid it, uh, was because obviously it's a parallel to the modern data stack, uh, which, uh, uh, you know, maybe we can talk about in a, in, in a minute, but like the modern data stack was both exciting, uh, but Um, also, uh, reflected a moment of excess. And, uh, you know, the modern AI stack may be taking the same general direction, uh, because there's a lot of companies, uh, this, uh, entire new categories are seemingly are getting crowded overnight. Uh, and, uh, you know, a lot of those companies, uh, to put it bluntly, uh, don't have a lot of traction yet, because, uh, it turns out that, If you're trying to do something, and I can, I'm not picking on them, but like evaluation or monitoring and all the things, well, you need to have LLMs to monitor, right? So you need, your customers need first to define what it is that they want to do, select an LLM, bring it in, and all the things, and then, you know, you can spend more time on the evaluation front, but, so there's a lot of companies that, um, Are getting started with this wave and, uh, that sometimes are a little bit, uh, hard to differentiate from, from one another. And, uh, yes, it's starting to feel a little bit …

AI assessment note: “the modern AI stack may be taking the same general direction”

Q you can try things really fast. Um, and then what is the bottleneck to coming up with new mathematical insights? It's gonna be new things to try and ideas to try. It brings up the question of like, um, how do you, how do you teach the next generation of people? How do you teach your kids, um, to become more creative? And I'm not sure how you do that.

A And then going, going back to the, Point you were making about, um, software. So indeed, um, this, this idea that all of this could be profoundly disrupting, uh, to the, I guess, what, what has been known as a SaaS industry. Uh, and software in general, and, um, you know, just like a couple of, uh, tweets, um, I saw, like, as I was, you know, prepping for this, so there's one from PG precisely that, uh, uh, went very, uh, viral where he said, I talked to the CEO of a moderately big tech company who said they'd replace Figma with Replit. This surprised me because I don't even think of them as being in the same business, but he said Replit is so good at generating apps that they, uh, just, uh, Go straight to prototype now. So that was, that was one thing. And then, um, another, uh, tweet from, uh, Chris Branridge, who said that, um, he had just built a Typeform clone in 20 minutes for three dollars 50 using Repli's AI agent. Uh, so just a couple of examples, but, um, like, indeed, uh, you know, if you can build any software, then what does that mean for entrepreneurship? What does that mean for venture investment? Uh, does any company have any, uh, moat, at least from a technical standpoint? Uh, and then, you know. Is that a good thing? Is that a bad thing? It's, it's so huge as a concept that it's hard to wrap your mind around it.

AI assessment note: “going back to the Point you were making about, um, software”

Q the point that you were just making about, you know, what happens to some of these other open source players that don't have the luxury like Meta do of maybe, you know, just open sourcing to open source for a power play and do need to make money off of it. So what happens to commercial AI, um, in the wake of kind of this explosion of open source development?

A Yeah, very much so. And, uh, a really interesting question is, you know, for now the, Open source model are not as highly performant as the commercial models, but, you know, maybe there's a world where they do become as performance and maybe even more performance. So what does that mean in terms of like overall, uh, sort of, uh, distribution of, of value and, and who grabs the value where in the industry? That's, that's going to be another, uh, part that's going to be fascinating to, to watch. So, you know, looking in the meantime for the commercial LLM Players, um, you know, it's been like everybody's favorite question, given the enormity of the amounts, uh, raised in a very short period of time by all the, uh, LLM players. Like, are we, are we witnessing this crazy incineration of, of capital? Uh, you know, people have called, uh, LLMs, uh, I think the expression was the, the, the fastest depreciating asset Uh, in history. There's certainly, um, those are certainly interesting and, and valid points with, with a lot of, uh, you know, strong rationale behind them. Uh, but equally, if you look at, uh, OpenAI and Ananthropic, uh, they're doing very well. Thank you very much. You know, they're just, uh, growing at, uh, astounding pace in terms of, of revenue. So there's a little bit like that, like cognitive dissonance and what, obviously what is true Now may not be true in, in, i…

AI assessment note: “If your product is the LLM itself, then, you know, maybe that's tricky”

Q Did you also try to generate the answers?

A Yeah. And the video, and the audio, and then, uh, and then we can all focus on the pizza here. Do what we would do best, which is eating. Um, that's, that's, that's a great idea. Um, so one of the questions was, uh, what, what you just started getting into. How does Perplexity AI certainly differ from those of Google and Microsoft? And what unique features does it offer? And I'd love to jump into this great question from Perplexity AI, um, and have you talk, maybe give us a little bit of a product tour. So that's the core product. Um, and then there's a pro version and then there's perplexity labs in the API products. If you could take the three in turn, that would be great.

AI assessment note: “Yeah. And the video, and the audio”

Q a fundamentally very hard problem because, you know, when you connect in an unlimited way, you can break things in the real world. And we don't like models to break things for us. So that's the part where people work a lot. It overlaps with security, right? You need to like have very good security to allow models to go on and train on the things they need to train.

A One theme that people like me, VCs and, uh, you know, founders and startups think about a lot as we see all the progress at OpenAI is as the models keep getting more general with more genetic capabilities, the ability to run for a very long time, you know, going to areas like science and, uh, math and, you know, recently it was reported that, um, there were, uh, some, uh, Uh, investment bankers hired to help improve the model's capability to do grunt investment banking, uh, work. All of that taken into account, uh, is, is there a world where basically models, or maybe just one model does, uh, everything? I don't know if that's AGI, let's not necessarily go into that debate, uh, but what's left for, uh, people that build products that sit on top of models?

AI assessment note: “what's left for, uh, people that build products that sit on top of models?”

Q But it's also not a game entirely, because it doesn't follow the rules and the instructions that we know of games. So what is it? I don't know. Um, I know it's just different. It just feels and tastes and smells different. Um, and I think real time is probably one of the things that will unlock many other new use cases that just were never manageable for people before.

A Are you a believer in the same vein in the, this idea of a hyper personalization, uh, of content as in the movie format? So maybe not as, as crazy as what you just described. Uh, but, uh, you know, I like it. tweeted a while ago this idea of, um, That you could, um, uh, have a Netflix series that would be just completely based on, I want this actor doing this thing in this scenario, and, um, I got a lot, a lot of flack for it, like, people were not happy, uh, for whatever reason. Uh, did you, is that something that you think is possible? Is that something that you hear people in the creative industry, um, uh, talk about?

AI assessment note: “maybe not as, as crazy as what you just described”

Q around like, oh, you do like both, you train the model, but you also, you run the model at the same time, and that they have to happen at certain cadence versus of pre-training the seeds of distributed, uh, pre-training. Like you had that in TensorFlow, which Google released in 2017, right? So there is, it's a much more mature stock, uh, versus what you need on the post-training side.

A All right. So maybe to, as a last part to this, uh, conversation, it's been, Really fascinating and illuminating everything that you guys have described, because in particular, it sort of highlights the complexity of the systems, like the multiple stages. And I love what you mentioned, Nathan, a few minutes ago when you said the pre-training is scientific and post-training. My words, not yours, but my interpretation of your words was like, it's a lot of tinkering and putting things together. In a way that you hope is going to, to, to work and, and, and truly diving into how those models work on the one hand, but on the other hand, you know, each time you're, uh, like open a newspaper online or go on Twitter, like everybody's talking about, uh, AGI and how we're almost there and how it's going to change everything. There's, there's a little bit of a cognitive dissonance between like the, the, the reality of, uh, trying to make those models work with all the, uh, Unbelievable progress that we've seen, of course, but like, you know, that on the one hand and the discourse on the other hand, Nathan, you've, you've, you've had a much more, I would say, tempered view of AI progress compared to some AI researchers. You had a great blog post very recently that you called Thought on the Curve. I'm curious what your latest Thinking is, and look, obviously feel free to jump in any time, bu…

AI assessment note: “So maybe to, as a last part to this, uh, conversation”

Q versus a fight for like contracts. Cause in the past it was like, okay, you sign a ten million dollar contract for Teradata. You're kind of stuck with it, right? Like there's no choice now for the next three to five years. Uh, now it's like, well, you could use Snowflake, or you could use Databricks. It's just whichever one convinces you, uh, it's better for whatever workflow you're doing.

A You know, that, uh, I saw you, you, uh, write in one of your, one of your posts, uh, which I thought was super interesting, especially, like, in contrast to, like, this whole, you know, titanic fight between Databricks and, and, and Snowflake is that, like, a bunch of the project that you worked on had to do with, like, SQL Server and on-prem migration. So, While those two juggernauts fight it out, uh, it sort of feels like there's a chunk of the reality where, like, people are like, oh, well, yeah, this whole cloud thing could be interesting for us, and is that, is that, is that fair? Is that what you see sort of in the, in the trenches?

AI assessment note: “I saw you, you, uh, write in one of your, one of your posts”

Q have all these different types of data. How do you connect them to each other? If we have information about, you know, where people are, what they're spending, what the demographic is, how does that map back to the type of company we're looking at? That's probably much more exciting than, you know, there's 50 cars in the parking lot today, but tomorrow there's a cloud and they couldn't count.

A And, um, Um, you hear when you, especially for somebody like me in venture capital, when I, you know, I speak to a lot of startups, and, um, especially over the last two to three years, there's been this growing excitement to the point of maybe, you know, since borderline mythical about, hey, um, you know, we have, uh, we don't really have a business model, but we build this fantastic data asset, and, uh, you know, some hedge fund's gonna turn around and buy it for, like, ten million dollars a year. Um, which, um, you know, initially made me smile, and still makes me smile to, to, to, to, to, I mean, a supportive smile. Um, but, um, uh, at the same time, I actually have heard cases where that is true. Um, so what, what's, uh, is that, is that just like a one in, um, you know, thousand or million type occurrences? Uh, how should startups think about selling data to hedge funds?

AI assessment note: “How should startups think about selling data to hedge funds?”

Q But in theory, right, like you could hope that, like, in the future you could write an LLM that, um, what do you call it? Could write the API connectors better. Cause that's usually the big thing. It's like the API connection tends to be the one that is the hardest.

A Actually, taking a step back as, you know, we, through this whole discussion, you know, there's a lot of conversations about, you know, a lot of part of the conversation about sort of complexity, lots of tools, and all the things, and, you know, as you know, like I do every year, this big market map called the Mad Landscape, where there's like every year like a more, you know, even more gajillion, you know, new tools kind of thing. Why is it so Complicated. What, why, from your perspective, why are they, why do you need to put together those, like, chains of, like, tools, and sometimes it works, sometimes it doesn't. You mentioned, like, broken pipelines, that seems to happen all the time. Like, why are we here?

AI assessment note: “Actually, taking a step back as, you know, we, through this whole discussion”