Comment by fsddfsdfssdf

Conversations in its training data that explicitly mentioned "make no mistakes" don't strike me as particularly rich sources of high-quality reasoning signals. They strike me as conversations with Pointy-haired Bosses.