GTC DC ’25 Pregame – Chapter 1: State of AI Innovation

0:00:16 Hello, and welcome to a special GTC edition of the NVIDIA AI podcast. We’re dropping five
0:00:22 episodes on the road to GTC Live in Washington, D.C., bonus conversations you won’t hear anywhere
0:00:26 else. This first one dives into the state of AI innovation. We’ve brought together leading
0:00:32 investors and founders to examine how new ideas, models, and open collaboration are shaping the
0:00:38 direction of AI and where the next wave of durable innovation is coming from. Enjoy the conversation
0:00:45 and check out our catalog of episodes when you’re done. Just visit ai-podcast.nvidia.com. Now on to
0:00:51 the episode. Behind every breakthrough in AI are the innovators and builders turning possibility
0:00:56 into progress. From the open models to agentic systems, they’re accelerating the next wave of
0:01:03 innovation across startups, labs, and markets. Joining us to explore the state of innovation,
0:01:10 some of the best. Thomas Lafont, co-founder of Cotu Management. Sarah Gual, founder and managing
0:01:19 partner at Conviction. Martin Casado, general partner at Andreessen Horowitz. And Naveen Chadha, managing
0:01:28 partner at Mayfield. So Thomas, welcome. You know, Cotu has been investing at the heart of America’s
0:01:36 super cycles from the internet to cloud to social. Help us contextualize how AI stacks up and how you think
0:01:43 of all this investing in the middle of all the bubble talk. Yeah, good morning, everyone. I’m really thrilled
0:01:52 to be here and to be the warm-up show. So I know we have a lot of exciting content today. So look, I think
0:01:58 that all of the investments have really kind of been focused on infrastructure. So I think when people talk
0:02:06 about AI, obviously it starts with semis, it starts with power, it starts with the large language models. I think, you know, if you look at the private
0:02:11 private markets as an example, most of the value has been accrued at the infrastructure layer for the past
0:02:16 kind of five or ten years, let’s say. But to me, what’s most exciting about the moment that we’re
0:02:24 at right now is that we are seeing a lot of value be accrued at the application layer. And a new class of
0:02:32 companies start to emerge in different verticals. So as an example, if you look at coding, Cursor has become an unbelievable breakout
0:02:41 company, one of the fastest growing companies ever. Martine can talk about it as well. That’s only enabled to make coding
0:02:47 better because of the investments in infrastructure, right? So starting with NVIDIA. But we’re also seeing it in medical with open
0:02:54 evidence. We’re seeing it in legal with Harvey. So to me, one of the interesting moments that we’re at right now is all of the
0:03:02 investments in infrastructure have enabled apps to come through that are delivering real productivity gains, which I think then reinforces the
0:03:09 belief that it’s worth making these infrastructure investments because you can see the return from these applications.
0:03:11 Well, well said.
0:03:16 Martine, you know, you’re one of the legendary software investors in Silicon Valley.
0:03:43 You know, Satya was on my pod last year and made some news when he said, you know, AI is potentially a real threat to software. The software may be this thin interface on top of this CRUD database. And so we’ve seen a lot of, I think, FUD there. You know, talk to us a little bit about how you expect software to change, what’s winning and what potentially loses in the age of AI.
0:04:13 Yeah. So I think formal languages came out of natural languages for a reason. And that is natural languages, you can’t actually describe what you want. Right. So we’ve always had kind of two stories of software, right? We have the low code, kind of drag and drop. And that is definitely be transformed by AI. I think that’s gonna look entirely different. But then we also have the very highly technical where there’s actual trade offs, you have to understand those trade offs. And, you know, you still need professional developers. If you actually look at AI,
0:04:41 the primary use right now in development is professional developers. It isn’t kind of casual, if you do it dollar weighted. And so, listen, I think we’re in for a major transformation. I think we’re going to see a lot more software than we had before. I think many more people will be able to develop than they’ve ever been able to develop before. I think it’s a great educational tool, but I don’t believe it’s going to get rid of software development. I think this is very much a technical discipline where you have to understand the trade offs to do it. And so, but I will say, I mean, I’ve been in software for 30 years.
0:04:48 30 years, and this is the first time we were being disrupted, right? And so definitely is putting us on our heels to try and understand what’s going on.
0:04:51 Yeah, makes sense. Daveen, great to see you.
0:05:08 Absolutely. So you’ve predicted, made a bold prediction out there that knowledge workers will have what you’re calling AI teammates. And that’s a $6 trillion opportunity. And I don’t know if that’s by 2030 or what timeframe this is.
0:05:16 Can you tell us about what an AI teammate is and how it represents a massive way that we shift in our work?
0:05:29 Absolutely. So our belief is AI is going to team up with humans to get us to superhuman level. And we are entering an era of collaborative intelligence.
0:05:48 What’s going to team up with humans, what’s going to happen is AI will manifest itself in the form of teammates, which are nothing but digital companions that team up with us, not only to accelerate productivity, but augment our capabilities and amplify our creativity.
0:06:04 If you look at globally, the knowledge worker spend is $30 trillion. So if AI takes 20% of the market, five years, 10 years, it’s a $6 trillion opportunity. First time, we’re not going after IT budgets.
0:06:21 This spend is coming at team, 10% of the people spend for the knowledge workers. And same is going to be the case for physical AI. So we are extremely bullish. It’s the same size as the IT market. Now it remains to be seen. Is it five years, 10 years? But it’s going to happen.
0:06:41 Let me ask a quick question to follow up to that. Is this deflationary at the end of the day, right? Does the $30 trillion market, because we’re a lot more productive, right? So the idea is replacement, that $6 trillion will go from humans to machines. But might it also just offer some deflation to the economy, some productivity gains along the way?
0:07:03 Absolutely. So I think in any new information technology market, there is displacement because productivity gains happen. There is some job loss. But in the long run, I’m an optimist. Humans come out as winners. The more productivity gains you get, the more cost savings you get, the more profits you create, you’re going to hire people.
0:07:33 By the way, the difference is, basically, they won’t be doing mundane jobs. They’ll be doing things that weren’t possible before. And one example we talked about, the white coding, 30 million developers have been able to code. So creation has been limited in the form of company creation to 30 million people. Now with white coding, which is a teammate, a billion people can become creators, and it democratizes entrepreneurship, and we can’t even figure out.
0:07:43 What people are going to do what people are going to do with this technology. So I’m very bullish. Short term, they’ll be paying, but long run, the avenues are infinite on what gets created.
0:07:57 It’s a really important, we’re in Washington, D.C. And at a time where I think there’s a lot of confusion on Capitol Hill about the impact of AI. We have too many doomers running around talking about how it might shrink the economy, right?
0:08:11 People are going to be displaced, jobs will be lost. It’s important for all of us, and everybody’s going to come after us, to come here and educate really on the abundance to come. America cannot stay competitive unless we have our best innovators leading at the front, right?
0:08:18 That starts with federal preemption on state laws and other things that can happen here in D.C. in order to accelerate AI. Sorry to interrupt.
0:08:32 No, not at all. Listen, I’m a student of history. And what history does show is with every major inflection in technology, whether that’s engines, electricity, it has been replaced.
0:08:40 Now, this curve is dramatic. I think more dramatic than we’ve seen. And the other ones, we do need to keep an eye out for it.
0:08:42 Sarah, how are you?
0:08:44 I’m great. Thanks for having me.
0:08:54 Absolutely. So you’re an AI-focused, AI-native shop, which is amazing and a pretty good place to be right now.
0:09:02 Let’s talk about Open. I’m curious about how does Open drive, let’s say, growth?
0:09:09 I think we know where growth is, but things like national security, how the U.S. can win by adopting Open.
0:09:16 Okay. So I think there’s two really important questions here. There’s Open as an open source, right?
0:09:30 As an engineer, but even for everyone who isn’t, like the foundational principle of open source is that if you allow more people to participate in innovation,
0:09:39 you will get better ideas, you will get compounding innovation, and then importantly, you will open up the market at the layers above that innovation, right?
0:09:51 And so I think the idea of open source and open models in particular allows for more democratization at the application level where more entrepreneurs can build things that get to actual end users.
0:10:01 Thomas mentioned, you know, Harvey and law, open evidence in medicine, cursor, Martin mentioned in engineering, we’re like 1% of the way in, right?
0:10:06 What about every other job and vertical? Where are the tools for that? And I think open innovation is going to allow for that.
0:10:17 You know, on the national security front, my view as a big believer in America is that we have always won by being strategically open and at the forefront, right?
0:10:30 And so I think these things can go together. Strategically open means like, to me, attracting the capital and the talent to create the technologies that lead and create abundance.
0:10:39 Strategically open is deciding what pieces of that we really want to own, both in a supply chain perspective, and then what companies really matter and will create opportunities for Americans.
0:10:46 Yeah, it really seems like we should be embracing, at the model level, all models, regardless of where they come from.
0:10:50 And of course, we can go in, we can expect the results to see if there’s something going on.
0:11:02 But I do think that there’s the meme that says we shouldn’t be importing models from other countries, namely China, to be able to drive innovation and stacks on top.
0:11:03 And I do think that’s a mistake.
0:11:14 I think that when I look at it from the application developer’s perspective or the end user’s perspective, they’re going to choose tools that are good for their tasks, right?
0:11:18 I don’t think you’re choosing like a philosophy of intelligence.
0:11:23 You’re looking for, honestly, efficiency and capability for most of these things.
0:11:27 And so I think efficiency and capability can come from all corners of the world.
0:11:34 And let’s be honest, you know, if it’s driving American innovation, let’s have a little confidence in the entrepreneurs that are adopting them.
0:11:34 That’s right.
0:11:48 You know, and so we had this DeepSeq moment earlier in the year that I think, again, led people to believe that we can live in this artificial world where we shut down all the open innovation in China and just drive it in the United States.
0:11:58 And while I’d like to see, and it’s been great to see all the open source models coming out of U.S. labs, the fact of the matter is DeepSeq accelerated innovation in the United States.
0:12:00 I don’t know, Martin, do you have a…
0:12:04 So, listen, I think not having a policy is potentially dangerous here, right?
0:12:14 So, let’s imagine that models were 3D printers and we allowed anybody to adopt any 3D printer that they wanted and we decided to limit our own 3D printers.
0:12:20 So, if our entire manufacturing foundation is based on somebody else’s 3D printers, there’s a lot they could do, right?
0:12:24 They could, you know, modestly shift what comes out.
0:12:29 They could only release weaker 3D printers for everybody else and they keep the stronger ones to themselves.
0:12:34 So, when it comes to import controls with technology, we’ve had longstanding policies, right?
0:12:37 You remember, like, the whole Huawei Cisco thing in the early 2000s.
0:12:39 And, listen, it turned out to be very appreciate, right?
0:12:46 We shouldn’t run our critical infrastructure that we rely on on something controlled by a foreign adversary in every case.
0:12:49 And so, I would say, first, it’s important to have a policy.
0:12:52 That policy should be nuanced to understand everything that you said.
0:12:53 Yes, we do want to benefit from it.
0:12:55 Does it go all the way down to critical infrastructure?
0:12:56 Maybe, maybe not.
0:12:59 Historically, we’ve done a pretty good job not doing that.
0:13:06 And so, I think I would recommend being a bit more thoughtful of how we import and use open source areas.
0:13:08 Naveen, do you have a comment on this?
0:13:10 Yeah, I think I agree with what was said.
0:13:21 But having been in the industry for a long, long time, my belief is the reason openness becomes important is it’s an ecosystem opportunity.
0:13:24 No one company can solve all the problems.
0:13:36 So, we have seen with prior platforms, whether it was Windows, Android, iOS, you need a whole ecosystem that thrives and just keeps innovating.
0:13:44 So, it’s not only the geopolitical situation, but for costs and the problems to get solved, you need openness.
0:13:49 And I’m a big believer what happened in the past is going to happen again.
0:13:55 And companies which are platform companies, which enable ecosystems, will solve problems together.
0:13:58 So, it’s a together thing rather than one company doing it.
0:14:01 Thomas, I want to come back to a question I started with you.
0:14:10 Probably the number one question I get asked in venues like this is, how can you invest when every day on CNBC, they’re talking about a bubble?
0:14:14 And, you know, we have had to invest through all of this talk.
0:14:20 So, maybe just help us understand, where is KOTU investing the most today?
0:14:23 Public markets, private markets, early stage, late stage.
0:14:31 And how mentally do you guys continue to forge ahead while all of this chatter is going on?
0:14:51 Yeah, I mean, I think if you look at the public markets, for example, one of the different features of this particular market versus the market in the 2000s, which, by the way, we were investing then, we started our business in 1999, is, first of all, stocks are valued very differently today than they were back then.
0:15:06 So, to me, one of the incredible things about this market is you get to buy some of the world’s best, most exciting, most innovative, best-run companies, companies like Meta and NVIDIA and Google and others, right?
0:15:13 At P, multiples that are on average in the, call it mid-20s on a forward basis, depending on which company you’re looking at.
0:15:18 Still pretty incredible to get access to all of that innovation at that price, right?
0:15:25 So, but obviously, what we got to make sure is that we’re right about the E, right, in earnings, not just the multiple.
0:15:31 So, what we look for is leading indicators like ChatGPT, as an example, right?
0:15:34 And we’ve seen just off-the-charts usage.
0:15:48 One of my favorite things that Jensen talks about is the triple exponential that ChatGPT is benefiting from, which is more users making more queries and more deep research per query, right?
0:15:50 So, you kind of get this triple exponential demand.
0:15:55 So, I think watching ChatGPT and how it performs across the globe is really critical.
0:16:00 But, you know, you asked the question about inflation, right, or deflation.
0:16:07 To me, one of the defining features of AI is it actually has a chance to bring deflation to sectors that really need it.
0:16:10 So, I think we’ve looked at healthcare, for example.
0:16:16 You could argue that the internet really didn’t do much to bring healthcare costs down, right?
0:16:18 In fact, they’ve continued to increase.
0:16:22 I think AI really has the potential to dent the cost curve, right?
0:16:28 Companies like Open Evidence, as an example that we’ve talked about, that make the diagnosis much better.
0:16:30 Industrials.
0:16:32 How do we bring industrial back to America?
0:16:34 Well, we’ve got to get more efficient.
0:16:35 We’ve got to get more out of people.
0:16:36 We’ve got to get more out of machines.
0:16:37 AI can do that.
0:16:41 So, to me, all of those sectors, we’re seeing it in defense, right?
0:16:45 Look at new companies that are coming in with lower-cost technologies.
0:16:48 I think we’re going to see that kind of across the globe.
0:16:50 All of those things kind of give us the confidence.
0:16:58 That said, and there’s always a but, we do think hypervigilance is really the key.
0:17:00 So, we do watch these positions very carefully.
0:17:03 We look at all these leading indicators very carefully.
0:17:06 I don’t think it’s a time where you can be kind of complacent.
0:17:07 Right.
0:17:10 So, Martin, I’ve got a question for you.
0:17:11 We’re sitting here.
0:17:12 We’re talking about AI stacks.
0:17:26 And just when it’s gone from the GPU to a tray to a full rack, all the software, and now we’re talking half the time about how we’re going to power these, how are we going to build all these?
0:17:34 Where is the leverage and the most leverage in the AI stack right now?
0:17:41 I mean, honestly, let’s just say from a Nashville standpoint, we wanted to increase AI throughput.
0:17:42 And there’s one thing you can do.
0:17:44 It turns out that’s not technical.
0:17:50 The one thing that we can do is ease regulations on breaking ground for new data centers that power.
0:17:51 Full stop.
0:17:51 Yeah.
0:17:54 I mean, I think that the rest of the stack, we understand very well.
0:17:57 That right now is what is limiting our ability to do massive capacity buildup.
0:18:04 By the way, Martin, I’m sure you saw yesterday, but OpenAI wrote an open letter to regulators.
0:18:04 Yeah.
0:18:07 Probably worth kind of touching on today, right?
0:18:14 But essentially calling for a Manhattan-like project, right, around power generation in this country, right?
0:18:15 And looking at all of these options.
0:18:19 And look, they called for 100 gigawatts per year, which is a massive number.
0:18:22 But I think directionally, they’re correct, right?
0:18:24 Chips need power.
0:18:27 And to Martin’s point, we need to invest in it.
0:18:28 Yeah.
0:18:31 I think we often underestimate what all of this means.
0:18:33 We throw around terms like gigawatts all the time.
0:18:35 Every time I’m in a picture, I need a gigawatts data center.
0:18:38 We’re talking like four football fields worth of capacity, right?
0:18:43 So this is a major national level undertaking that we need to do.
0:18:46 Definitely need public-private partnership and cooperation to do that.
0:18:51 But if there’s one thing, one thing that we can do, it’ll be regulations on building our new data centers, especially power.
0:19:00 I do think it’s also worth noting that there are very few people who believe that America could not benefit from an updated grid.
0:19:01 Yeah.
0:19:01 Right.
0:19:11 Every part of it, transmission, storage, new generation capacity, and that there’s going to be a surplus to consumers from all of that if it’s invested in.
0:19:21 And so the fact that there are large buyers of that power that want to front the CapEx to that improvement in American infrastructure, I think is a really good thing.
0:19:27 I mean, I think, you know, Gurley and I did the pod from Diablo Canyon, the nuclear site in California.
0:19:33 That fortunately, I mean, amazingly, they were talking about shutting down.
0:19:40 It represents 12% of the clean energy, 12% of the total energy in California, vast majority of the clean energy in California.
0:19:41 We’ve gotten that extended.
0:19:45 But the reality is the site is situated for four new reactors.
0:19:54 And we suggested on the pod that Meta, Google, OpenAI, Microsoft could all, you know, build their own reactor with a data center right next to it.
0:19:58 Those are the type of out-of-box thinking that we need to have in the United States.
0:20:01 There are 100 fission reactors under construction in China.
0:20:03 Today in the United States, we have zero.
0:20:03 That’s right.
0:20:03 Right.
0:20:10 And if power is the primitive, if it’s power in and tokens out, then we got to get really serious about power.
0:20:11 And it all starts here in D.C.
0:20:17 I would also say, right, like we’re in the golden era of the semiconductor industry.
0:20:33 Beyond the GPU and accelerated computing, not only do you need these new supply of energy, but you can do amazing things on the power and cooling side with innovation that hasn’t happened for 40, 50 years.
0:20:39 So what goes onto the board with air cooling, liquid cooling, you can do amazing things.
0:20:43 And at the end of the day, the voltage and the loss you end up getting.
0:20:48 So there is like just cutting edge companies being created that weren’t happening 50 years back.
0:20:53 So it’s not only like supply of energy, but it’s also it’s being lost.
0:20:59 You know, speaking of primitives, Martine, one of the primitives is data.
0:21:04 And, you know, you, you know, been one of the pioneers around the modern data stack.
0:21:08 Of course, Databricks being one of Andreessen’s large investments.
0:21:18 Help us understand, you know, is AI a threat to, you know, data software companies, traditional database companies, the snowflakes and the Databricks?
0:21:23 Or are these things essential ingredients, you know, into superpowering AI?
0:21:30 I mean, I just, I think the best mental model for models is that they’re data frozen in time.
0:21:39 I mean, and you need a lot of machinery and plumbing to get data from the source, which is tends to be the natural universe to these models.
0:21:41 So if anything, it’s been a dramatic accelerate.
0:21:44 That said, you’re seeing kind of a tale of two worlds, right?
0:21:50 There’s like the traditional analytics data stack, which is all structured and not very AI ready.
0:21:55 And then there’s the new stuff, which is you just kind of throw a bunch of data at the model and see what comes out at the other end.
0:21:58 And so I would say, A, it’s a massive accelerant.
0:22:03 However, if you’re working on data, you need to kind of catch the shifts that are going on.
0:22:08 Because listen, it’s data using a new way with, you know, different guarantees and different bounds.
0:22:10 I have a related question here.
0:22:14 So the GSIs have literally millions of employees.
0:22:21 And typically what they do is they’ll do digital transformation projects, but they’ll also modernize, right?
0:22:23 And today, modernize.
0:22:28 Let’s say I’m going to take 27 instances of SAP and one.
0:22:33 Do you see a future with agents that would be able to do this?
0:22:40 Because where a lot of the training comes from on those is from PDF files and training classes.
0:22:41 It’s almost like knowledge bases.
0:22:48 Do you see agents as a potential driver to do that with software?
0:22:50 I’ll say that.
0:22:52 I would actually love to hear Sarah’s view on this.
0:23:09 But here’s my experience looking at a lot of AI companies, a lot of companies adopting AI, which is, let’s say, 80% of your time is branch work and 20% of your time is something else that actually involves agency, like knowing what the business needs or knowing what you want or knowing what the customer wants or whatever it is.
0:23:11 AI is pretty good with the 80%.
0:23:12 It’s horrific at the 20%.
0:23:13 It just really is.
0:23:15 And this doesn’t matter what it is.
0:23:16 So sure, you can do document processing.
0:23:25 But my experience over time is, you know, the AI will come, make you more productive on the stuff that, and be able to focus on the stuff that really matters.
0:23:27 And it will increase productivity.
0:23:30 And I think we tend to conflate this with things like job loss.
0:23:37 We’re really, we’re coming out of 2021, which is this kind of crazy time where you’ve seen like a massive compression on value.
0:23:39 And there’s retooling and skill sets.
0:23:44 Like the AI companies I sit on the boards of are hiring like crazy.
0:23:48 If that isn’t an indication that you need people and not AI, I don’t know what is, right?
0:23:51 And so listen, clearly there’s a shift in skill that’s happening.
0:23:53 That’s very important for us.
0:23:55 You know, clearly, you know, there’s some kind of macro stuff going on.
0:24:02 But my view is these things, yes, will take the drug work to your question, but they will drive human productivity, not replace it.
0:24:04 I mean, I think that’s such an important point.
0:24:09 Again, I was on Capitol Hill yesterday having some of these conversations.
0:24:13 There was news yesterday that Amazon is having, you know, a major riff.
0:24:17 And, of course, the question was, is AI causing them to lay off all these folks?
0:24:23 And what I reminded them of is coming out of 22, which we described as the age of excess, right?
0:24:26 In COVID, companies were hiring like mad.
0:24:28 They thought everybody was going to stay home.
0:24:32 Getting flatter and getting leaner and getting fitter, right, was important.
0:24:33 It had nothing to do with AI.
0:24:40 The most dangerous thing about AI, in my opinion, is that it’s an excuse to do slimming, not that it’s actually changing productivity.
0:24:45 So I think it’s important, you know, that Amazon came out this morning and said,
0:24:53 while we are going to get leaner in order to get more competitive, it’s so that we can hire more people and double down on our bet in AI.
0:24:57 One of the things I want to come back to, you know, we talk a lot about Anthropic and OpenAI.
0:25:07 Two companies that, you know, don’t get nearly as much airtime when it comes to models, X.AI and what they’re doing around physical intelligence.
0:25:22 And then Meta, you know, I think there were some, you know, commentary yesterday about, you know, Meta was kind of a surprise how little they accomplished on the model front over the course of last year, given their focus.
0:25:29 So, you know, Thomas, Sarah, I would love for you guys to talk to us a little bit about the other models that exist out there.
0:25:34 Were you surprised that Meta didn’t have a bigger impact with Llama 4?
0:25:44 And where do you think, you know, the next wave comes from on X.AI, given, you know, that Elon’s out there teasing that he may be the first to AGI?
0:25:49 Yeah, maybe I’ll talk Meta and then I’ll listen to Sarah on the next generation.
0:25:54 I think when we talk about Meta and AI, there’s a couple just to kind of level set, right?
0:25:59 And the first thing is LLMs are broadly not in use at Meta today.
0:26:11 So if you think about the family of apps from Big Blue to WhatsApp to Instagram and threads, LLMs functionally outside of small features and threads are not in use today.
0:26:22 So I don’t think that the bet that Zuck is doing is about, frankly, today, or might not even be about winning the desktop agent war, right?
0:26:25 Let’s even presume that maybe ChatGPT has kind of won that battle.
0:26:29 So why is he choosing to invest so aggressively?
0:26:37 Well, I think what they see is a world where one day LLMs are going to be in use inside of all the big apps, right?
0:26:41 So you will see generative AI ads, right?
0:26:45 Ads that are kind of customized to the individual user generated on the fly.
0:26:50 So we are going to see tremendously more LLMs in use inside of those apps.
0:26:54 And they probably want to make sure that that’s run on their technology, not someone else’s.
0:26:58 So I think that’s where the investments kind of come from.
0:27:05 It isn’t just kind of chasing wanting to be a player on the kind of the assistant side, right?
0:27:12 But it’s realizing that AI is going to be a core part of all of the infrastructure of their core products over the next decade.
0:27:15 That’s the time frame I think we should kind of judge them on.
0:27:20 They want that to be run on their own technology, but they’re also capitalists.
0:27:22 And if their own technology can’t keep up, they’ll turn to others.
0:27:25 So maybe you can tell us about what you’re seeing from the others.
0:27:36 You know, I think your view on XAI depends on whether or not you think this is like where you think we are in the model development war overall.
0:27:45 If the front is infrastructure, if the front is new architectural breakthroughs, or if it’s like capital raising, right?
0:27:48 These are all very reasonable assumptions right now.
0:27:53 A lot of people would say Elon knows how to build big stuff fast, right?
0:27:57 And navigate the regulatory and resource landscape around that.
0:28:01 If it’s infrastructure, I think X has a very good shot.
0:28:11 That being said, this is a period of time where you can’t trust the narrative from any individual company that much in AI.
0:28:14 You know, it ricochets all the time.
0:28:20 But there’s a period of time where there’s a lot of industry consensus among the leading researchers that scale was all you needed.
0:28:26 And that if you just put more compute into pre-training, you would get more capability out.
0:28:37 I think it’s pretty clear now that the returns to that scale is slowing down, and there might be more efficient ways to spend the next gigawatt of power if we can get it.
0:28:40 If that’s true, I think it’s a much more open landscape, right?
0:28:50 And you see really interesting companies like Thinking Machines and Reflection and these new labs staffed by amazing researchers making a new series of bets on different capabilities.
0:28:56 I’d also say ChatGPT is an amazing product benefiting from these three exponentials.
0:29:07 For consumers, whether it’s from ChatGPT or from new products, I think we’re still like 1% of the way there in terms of the experience that’s possible, right?
0:29:16 There’s still like we’re at the very beginning of multimodality, figuring out how to make reasoning like cheap and efficient so people actually use it.
0:29:21 Very few people use the latest models from OpenAI all the time or from any other vendor.
0:29:29 And I think the idea that we talk about agents and there are some instances of those being used in the business context.
0:29:33 They’re not broadly used by consumers today proactively.
0:29:41 And so I just think we’re going to see many more experiences where the landscape of competition is not set between Meta and XAI and all these others.
0:29:43 So it kind of depends on what you think about the research.
0:29:50 I have a question when it comes to ChatGPT and answers.
0:29:54 I think that’s where most Americans and most users interface with AI.
0:29:56 And it’s magic.
0:29:59 You pull out your phone and you get an answer to almost any question.
0:30:08 I feel like the next 10x moment is when that personal assistant can take actions, can book my hotel, can buy my black t-shirts, right?
0:30:11 Can do all the things that a great assistant can do digitally.
0:30:21 And so my question to you is, when is our agent, when is our ChatGPT going to be able to book our hotel for GTC?
0:30:27 Where I simply say, hey, Chat, book me the Hayes Adams for next Tuesday in Washington at the lowest price.
0:30:29 Are we going to see that in the next six months?
0:30:38 I think six months is tough, but it’s just, we’re going to blink and it’s going to happen because that’s the power of these models, right?
0:30:42 Like not only can they reason, plan, but they have to take action.
0:30:45 But there’s some missing things on the integration side.
0:30:47 No science problems.
0:30:51 These are all eng problems and it’s just around the corner.
0:30:57 And it is happening in the enterprise first because you can do manual work, you can do integrations.
0:31:06 So we are seeing a lot of cases where some of these things are able to complete loop and take action and take it all the way.
0:31:11 It is going to happen because at the end, as I said, we’re going to have multiple teammates.
0:31:12 We are seeing it in sales.
0:31:13 We are seeing it in legal.
0:31:15 We are seeing it in coding.
0:31:17 In the enterprise, it’s going to happen in consumer.
0:31:26 I mean, Brad, to me, the most powerful leap will be when it’s not just executing the idea, but when it’s actually generating new ideas for you.
0:31:30 And I think the Pulse product from ChatGPT is kind of a window into that.
0:31:34 So that’s going to be the kind of the really exciting part.
0:31:36 Well, we were lucky to have you guys.
0:31:45 For everybody in the audience, for the absolute best in venture and now leading the charge in AI, it was a thrill to have you guys here.
0:31:50 So that’s going to be the chance to have you guys here.
0:31:51 So that’s going to be the best.
0:31:52 So that’s going to be the best.
0:31:53 So that’s going to be the best.
0:31:53 So that’s going to be the best.
0:31:54 So that’s going to be the best.
0:31:55 So that’s going to be the best.
0:31:56 So that’s going to be the best.
0:31:57 So that’s going to be the best.
0:31:58 So that’s going to be the best.
0:31:59 So that’s going to be the best.
0:32:00 So that’s going to be the best.
0:32:00 So that’s going to be the best.
0:32:01 So that’s going to be the best.
0:32:02 So that’s going to be the best.
0:32:03 So that’s going to be the best.
0:32:04 So that’s going to be the best.
0:32:06 So that’s going to be the best.
0:32:07 So that’s going to be the best.
0:32:35 So that’s going to be the best.