Dwarkesh Patel

If you had fully solved continual learning drop out of nowhere, then sure, it might be game set match as Satya put it on the podcast when I asked him about this possibility.

But that's probably not what's going to happen.

Instead, some lab is going to figure out how to get some initial traction on this problem.

And then playing around with this feature will make it clear how it was implemented.

And then other labs will soon replicate the breakthrough and improve it slightly.

Besides, I just have some prior that the competition will stay pretty fierce between all these model companies.

And this is informed by the observation that all these previous supposed flywheels, whether that's user engagement on chat or synthetic data or whatever, have done very little to diminish the greater and greater competition between model companies.

Every month or so, the big three model companies will rotate around the podium, and the other competitors are not that far behind.

There seems to be some force, and this is potentially talent poaching, it's potentially the rumor mill in SF, or just normal reverse engineering, which has so far neutralized any runaway advantage that a single lab might have had.

I experimented with different video models to help me animate some of my essays.

But the thing is, my team and I are very opinionated about what we want the end product to look like.

And so for a video model to be useful to us, it needs to be able to follow our instructions for exactly what kind of shot and framing and lighting we want.

But all these labels, which would make it clear how to map from a specific style to a generated video, don't exist by default in the pre-training distribution.

So when one of Labelbox's customers wanted to improve their video generation model, Labelbox pulled together a team of expert cinematographers and editors and directors and had them annotate clips with concise technical descriptions so that the model would have context on things like dolly shots and Rembrandt lighting.

As you can see, unlocking broad economic value from these models requires not just coders and STEM PhDs, but people with taste and context in all kinds of different domains.

Labelbox can get you experts in all of the above and more.

So whatever skill you're looking to give your models, there's a good chance that Labelbox can help you.

Reach out at labelbox.com slash dwarkash.

So I've been trying to hire a writer for the podcast, but I ended up getting way more applications than I anticipated.

But as I started to go through them, I noticed that many of them didn't feel like they were actually written and submitted by real human beings.