@ArielKwiat IDK man presumably you can only interpolate inside the data distribution in pretraining, but wouldn’t be surprised if you could get novel behaviour in high temp RL
I haven't opened an IDE in a while. I've been using lazygit/fork to audit the changes agents have been making, but for various reasons it's suboptimal for my setup. I (claude/codex) vibed this. It might be useful for you too.
https://t.co/ehTcQ5C8Zg