‘Model commoditization’ is the next chapter in AI frontier competition: Bond Capital

Mention that deceleration from some of the frontier models actually that might cause concern, but I this could also mean that there might be some dislocations that create opportunities because it will create that one of those speed bumps that you mentioned that might happen sooner rather later. How does one play this then if you were in the markets or perhaps looking at opportunities because while there might be a slowdown perhaps a hiccup at the horizon that could create a reshuffling that might also create pockets of opportunities. How are you guys at Bond Capital perhaps viewing some of those pockets emerging if there are any? >> There certainly are and we think about AI in chapters and a narrative arc for how it's evolving. The prologue probably was this research phase where you had researchers at DeepMind and OpenAI building AI bots that could beat humans in board games like Go and video games like Dota. That transitioned to this ChatGPT moment where OpenAI launched ChatGPT to the masses and we suddenly saw AI in our hands and we realized this is a big deal. That then transitioned towards what I'd call the enterprise adoption moment where you had Anthropic take the helm as the most important company in the world as we realized the commercial value of AI was so enormous and it feels now to your point that we're heading into a new chapter where maybe it's not the frontier model dominance chapter anymore, but really a chapter around model competition, model agnosticism, maybe even model commoditization as you get price efficiency and enterprises finding more rational ways to use open source and other technologies to actually port over inference and intelligence at at better cost than what we had before. The other chapter that we're really focused on which is also very interesting and I'm sure you guys have talked about Muse on the show here is this idea of AI agents taking action as opposed to recommendations. And even though, you know, okay, Muse can deliver groceries. Who cares? It sounds kind of trivial. The the implications of an AI actually stepping out of just the recommendation engine that we've built our internet on of where do we go for flights or hotels or or food or restaurants into actually taking that action on our behalf does rewire >> But that's exactly where the rubber hits the road in terms of the regulation or the over-regulation, rogue AI agents etc. And we're seeing this play out in real time. As far as Open AI is concerned, then how strong is your investment commitments to Open AI with the IPO being set back somewhat and Sam Altman himself saying, "Look, we need to slow things down." >> Ultimately, what investors want, including ourselves, is a company that's innovating in a frontier capacity. >> Yes. >> And just because you're an incumbent or just because you have some academic moat, it doesn't give you some right to maintain that forever. As long as Open AI is actually innovating in a frontier and they're a frontier model, so they'll have to deliver increasing capabilities. The future is bright because there will always be a portion of the market that will pay a reasonable amount for high-quality and differentiated intelligence. That being said, for 90% of the white-collar work that we do, which is answering emails, filling out forms, writing emails, it doesn't require PhD-level intelligence, right? And so maybe for 90% of tasks that we do, cheaper models and and some that are actually asymptoting to the cost of the bare inputs, which are really just computing and electricity, we might have a world where intelligence is actually that free and that readily available. And in that world, the margin will not exist for 90% of intelligence and for basic tokens, it will it will exist for this frontier intelligence, which might only be a small sliver of the market. And then it will exist for these application layer companies that can productize this, you know, 90th percentile intelligence into bespoke ways that deliver value for customers. >> Because of the AI models in the US, because of the frontier and running so fast, they actually pose more of a risk of going rogue, whereas the Chinese agents are not there yet. But there will come a time perhaps, because of how fast these Chinese models are evolving, where this might actually be something they'll have to consider. Is perhaps the the question or the concern about regulating AI something that both sides will have to will and investors will have to be aware of. And perhaps even more importantly, can you regulate AI? Because it took us so long to get our heads around regulating Bitcoin and blockchain. This is like maybe five times, 10 times more complicated than what it What are your thoughts with regards to how you handle this very complicated situation as to what as to not just whether it's important, but how you're able to do it given how fast it's evolving right now? >> Yeah, I believe that frontier model companies have some degree of incentive to self-regulate. And we're seeing this already with some of the actions taken by Anthropic and OpenAI. It does not behoove them to have rogue models taking negative actions that are actually damaging their customers. And so the checks and balances are already inherently there even from a purely capitalist standpoint. Going forward the real answer is no one really knows how this technology is going to evolve. Even the researchers that we talked to at the frontier labs, they can't tell you with certainty, "Hey, we're going in this direction versus not." And so ultimately you have to take the position are you a techno-optimist or are you a techno-pessimist? And at Bond, you know, we really believe that humanity evolves. And 200 years we're all farmers. We could not imagine we're sitting here at these desks in front of computers and and no longer having to plow the fields, but here we are. I do believe that over the fullness of time we'll find interesting ways to leverage technology for the better and we'll improve our society as a result. >> And if we can turn to SpaceX where Bond is invested, the next big thing there is Elon Musk's vision of orbital data centers. Are they viable and could they be a reality? Could they unlock even further value? >> Yeah, on paper, if you can get these spaceships into orbit in an cost-efficient way, then yes, they are viable and the temperatures and a lot of the benefits of being outside of the scope of the US electric grid or the global electric grid actually do have advantages. The question is can you get these into space in a cost-efficient way? And that's an economies of scale issue, that's a technology issue, that's a supply chain issue. And over the fullness of time, do I believe they can get there? Yes. The time horizon of that and whether or not that will weather a capital market cycle or two, I think that's an open question. >> We constantly hear, David how the Chinese AI economy, whether it's the LLM developers, whether it's the physical AI developers, robotics, and the chips are creeping up. They are narrowing the performance, they're narrowing the innovation gap with their US rivals. Is that where we need to be then as investors? >> Investing in Chinese robotics companies. Yeah, there are comparative advantages to each geography and to each technology. And I do think it it appears right now that China does have some pretty meaningful advantages on supply chain side, from an access to the grid side, and from a robotic side as well. And so I do think that there will be some pretty incredible opportunities that emerge in China as a result of that. Opportunities that may not be as easy to access in the US as a function of the supply chain that exists here. China also has its set of challenges when it comes to investing in these companies in that it's a highly highly competitive market and when you have a consumer robotics player, any consumer phenomenon, the margins tend to compress very quickly as a result of the competition. It's very good for consumers. Living in China, you know, I was there a couple weeks ago and just walking through the streets, I mean, what a what an incredible place to be in terms of convenience and access to affordable technology. And so I do think there are huge advantages, but there are always challenges and opportunities in each geography. As investors, we're really focused on the US, that's where based, that's where, you know, we're headquartered in Silicon Valley. We've been there for 15 years and it's what we know best. >> Yeah, also no one does a price war in China, that's something we've learned over the last couple of weeks, David. But what do you think would change to make blood capital consider perhaps looking at China because the opportunities are there. What do you think are what what conditions would change that would make you reconsider perhaps do we venturing out there with for with regards to possible investments in that side of the AI sphere? >> It's not purely from a technology or financial reason why we are cautious on China market. There's also geopolitical and capital markets reasons as well. And I think as those stabilize and if we do evolve into a world where there's parity with the US then I think that could be interesting for us. But so far right now the US opportunity is so large so fast and also so global in nature. It's not like US companies are not selling in Latin America or Asia or Europe. We see enough opportunity there and it's what we know best and so we focus on that.