>when he new better

He was against LLMs, reinforcement learning, etc. Claiming that those are a dead-end. He was wrong. His own "JEPA" ideas have basically gone nowhere. Meta was right in moving on without him.