LeCun is not a skeptic of AGI; he's a skeptic that LLMs will achieve AGI. For what it's worth, I agree with him. I think we'll need innovations beyond just fine tuning LLMs with RL to take us to the point where they can reason about data they've never been trained on before.