We could definitely use RL to add blank, invisible "<thinking>" tokens whenever the model thinks it should. Or just allow it to say "Hmm." lol. Easy to test fr

We could definitely use RL to add blank, invisible "<thinking>" tokens whenever the model thinks it should. Or just allow it to say "Hmm." lol. Easy to test from a technical perspective is all I'm saying, and not a bad idea.