When a thinking/reasoning model encounters an easy task, something funny happens. The model finds correct answer/plan early, but then it keeps thinking and sometimes talks itself out of it. Overthinking, lol. It's trained on human data, after all.

→ View original post on X — @thatroblennon