How AI is opening up new markets and impacting the startup status quo with Sarah Guo and Elad Gil

**Sarah Guo** (0:05)
Okay, hi, listeners, today, you just have me and Elad shooting the, what's the appropriate term here? Shooting the breeze.
Yeah, but I want to start this shooting the breeze session by talking about this Goldman Sachs report that everyone's reading, which essentially says, I'm just going to get on my soapbox for a second here, that the title is something like calling the top on AI.
So for obvious reasons, I don't like it, but I do think it's worth decomposing for a second. I do encourage everybody to go skim this thing. So there's a bunch of interviews in it, and two of the core ones are from this guy, Darren Asmoglu and Jim Covello, the respectively MIT professor and the GS head of Global Equity Research.
And Darren is arguing, essentially, that AI is going to impact less than 5% of all tasks and the trillion dollars of CAPEX that people are spending on training models is a waste because AI will be unable to solve the complex problems. It's not built to do that. And Jim argues that, in contrast with the Internet, where you are disrupting something expensive from the beginning even early on versus having a very expensive solution that then becomes democratized. AI is very expensive from the very beginning. And then the other argument he makes is that any efficiency gains from AI will be competed away anyway. And so, like, none of the companies are going to gain from this. And so if we just, like, talk about Darren first, Darren's arguing about something he doesn't understand. Like, his claim is, you know, how do we know if scale works? More data won't make customer support reps better. I think, like, that's just a fundamental, like, misunderstanding of the technology and also objectively of what has happened over the last several years of scale and data improving capability and quality of model outputs.

**Elad Gil** (2:09)
I think a lot of these folks, too, by the way, are just kind of stuck in the old AI world. Like, I haven't read the reports, so I'm not talking specifically about these authors.
But a lot of people are treating this like old school ML, and they don't seem to realize that there's been sort of a breakthrough in terms of these transformer-based models or other architectures that effectively are both highly scale-dependent but also provide different types of functionality and features than, you know, you're sitting there and you're munging some data and effectively doing fancy regressions in some sense. So I think that's the other issue here in terms of a lot of what I hear. This happens a lot in healthcare.
You know, in healthcare, they always check about how data is the new oil. And you're like, data is not the new oil. You know, sometimes data is useful, and well-labeled data can be extremely useful, but, you know, a lot of it is also about the model and the application and everything else. And so I think there's just this broader misconception in terms of how this stuff works and what it means and all the rest of it.

**Sarah Guo** (3:03)
Yeah, I agree with that. I think this is actually a case of like this time it's different and also people lacking, you know, even the market state of what's happening on the ground. Like there are absolutely things that are cost effective to do today.
That's why you get actually a series of companies that are democratizing capabilities and really more on the prosumer side. But, you know, beginning to see things even in, for example, health care, traditionally really slow industry, where you go zero to five or 10 million of run rate in your first year. I think like, as you said, one sort of problem with this framework of thinking is, you know, you assume AI is like what it has been in the traditional ML world. The other is this assumption that the tech won't get much better fast and it won't get cheaper fast.
And I think the willingness to predict 10 years into the future of insignificant improvement is ludicrous when literally all of the people working on this tech are unwilling to, like you and I are probably unwilling to predict two years into the future, much less 10 years.

**Elad Gil** (4:01)
Well, I mean, the thing I'd predict two years into the future is that there's going to be even more broad spread applications of it. So I think it's almost the opposite thesis, which is this is the early days. And if you look at enterprise adoption of AI, most large enterprises are very early on. They think of it as three things.