According to @_avichawla, Kevin Murphy’s DeepMind overview links classical RL to LLMs with RLHF, PPO variants, world models, and multi agent methods.