New Breakthrough in Large Model Training! Meta Proposes LSP: Achieving Capability Enhancement Without Data

To address this, Meta has proposed a new reinforcement learning (RL) method called "Language Self-Play" (LSP), which allows ...