linksku

Did you try randomizing the experiments? Did you mean the base version or one of the crazier themes?

The initial variants were created by humans, then Claude used them as a starting point to generate more. You can see the data for each experiment. It'll automatically shut down the failing experiments and create new variants similar to the successful ones

Good idea, I can try different fonts

Seems like a decent idea if you look past the fact that the $1 million is called a "gift", it's gold, and it's Trump

If competitors like Deepseek need to distill to build similarly powerful models, that's mostly for evals right? I don't see how it can be cost-effective to distill for a significant amount of training data

What's GEO?