Hacker News new | ask | show | jobs
by huseyinkeles 470 days ago
I read somewhere which I can't find now, that for the -reasoning- models they trained heavily to keep saying "wait" so they can keep reasoning and not return early.