Can Open-Source Models Distill OpenAI To Death? | Greg Brockman

00:00And there's been this pattern where these massive models come out uh they get distilled uh by open source model makers and then open source is just a couple months behind the leading foundational models. Um and you know, I guess like when the investment was smaller, being a couple months ahead, uh you know, mattered a lot. But I'm curious now that the investment is so big um and the models are capabilities are increasing you know, fairly dramatically you know, uh as you go. Um how is this defensible in the long term if you're just going to have that pattern repeat over and over? Well, I look at it a little differently. Like I think that the real investment that we are making is into that end-to-end co-design, right? Of

00:44having a system a system of people, right? Who are producing this technology, right? A way of working together and some of this is about how you leverage these massive supercomputers to produce these models. Now, it is also the case that it's not as simple as you can take the output of these models and distill and you have exactly the model of the same capability. It's just smaller and can run fast. If that were the case, we would just do that and then we would also have a model that would be, you know, much more uh easy to serve in in many ways. And of course, there's a lot of art behind distillation. There's a lot of great things there. But the point that I'm getting at is that the real thing that

01:20we are investing in is the machine that makes the machine. Now, the at the deployment side, we think a lot about safeguards. We think a lot about mitigations and we do that for many many different aspects of how these models could be misused um in real situations. And that's something that we have been investing in for many years and we think about that across areas like cyber or thinking about that areas like bio that we have a I a long-standing effort that you can see in our preparedness framework, which is public, about how we approach these kinds of uses of the model and how we try to make maximize the benefits, mitigate the risks. And so I think it's a real motion that every piece of what

01:59we do needs to connect to the question of how do we continue to make progress, but also how do we make these models broadly available? Because that's something that we really believe in, that we believe this technology empowers people, and that we want it to benefit people and lift everyone up. Yeah, but just to go back on that, um the pricing on this model is, I think, double the last model, GPT-5.4. Um and so, from an economics or a business standpoint, the question would be, you know, let's say you keep on progressing, but because there's been all this infra- infrastructure that's been put towards training the models, if open source uh can deliver not as good performance, but almost as good, uh and do it cheaper, um

02:42how do you handle that threat? Well, again, I look at it a little differently. So, first of all, if you look at our history, which really is not driven by anything in competition, it's just like our our own sort of progress and and desire, we have dropped prices on the same level of intelligence year over year, sometimes by literally a factor of a hundred. Right? It's like, at least an order of magnitude year over year, sometimes literally a hundred. But, the thing that keeps happening, it's real Jevons paradox, where it's like you lower the cost of something, way more activity happens, right? And I think that what we keep seeing is that there are returns to intelligence, right? That for the kinds of tasks that

03:21these models are now capable of doing, that a little bit more intelligence goes a long way. And I think that is the story of 5.5, that in some ways, you can almost look at it as like, "Oh, there's just an incremental improvement in intelligence." But, I think there's going to be a massive improvement in terms of what people use it for. And by the way, I actually think that incremental is actually very much an understatement for this model relative to 5.5. You know, it's a point one improvement in some ways, um but I think that I that that actually really undersells the magic that we see within this model, and that the that our early testers have have really [music] seen in their practical work.