“We're on a bubble. I don't know when it's going to burst or how. I've said for a while that we're spending exponentially more money for a logarithmic increase in performance. There hasn't been a step change. Although one thing I did not predict: tool use did improve the models a lot. But from a strict model-quality side, it's maybe a linear improvement over a year, if even. And they're spending hundreds of billions of dollars now.”

give the two-second version, which is we're on a bubble. I don't know when the bubble is going to burst. I don't know how it's going to burst. I have said for a while that we're spending exponentially more money for a logarithmic increase in performance. I think you're seeing that firsthand. There hasn't been a step change in performance. Although, I'll be honest, and one thing I did not predict, tool use did improve the models a lot. And the ability to use- What do you mean by that? Just like web search integrations, the integrations that OpenAI has with, like, I, God, I love Codex. This is not part of the sponsorship here. I use it for literally everything. It integrates with Adobe Premiere. Like, I've used it to added videos. And because they've just spent so much time doing the computer use, the integrations, it does increase the power a lot. Now, from a strict API side and from a strict model quality and response side, it is a maybe linear improvement over a year, if even. They're spending hundreds of billions of dollars now. So I've had that thesis for a while. And then similarly, and this is, again, not something I necessarily predicted, but had something because open source has caught up a lot more, which…