Remix.run Logo
bonoboTP 15 minutes ago

They were RL trained on verifiable rewards. It's not purely learning to predict the next token of a human produced stream.