What do you make of the following exchange, which feels like a summary of his contrarian takes?
Lex Fridman (01:29:00) Well, so to summarize, you recommend in a spicy way that only Yann LeCun can, you recommend that we abandon generative models in favor of joint embedding architectures?
Yann LeCun (01:29:15) Yes.
Lex Fridman (01:29:15) Abandon autoregressive generation.
Yann LeCun (01:29:17) Yes.
Lex Fridman (01:29:19) This feels like court testimony, abandon probabilistic models in favor of energy based models as we talked about, abandon contrastive methods in favor of regularized methods. And let me ask you about this, you’ve been for a while, a critic of reinforcement learning.
Yann LeCun (01:29:36) Yes.
Lex Fridman (01:29:38) The last recommendation is that we abandon RL in favor of model predictive control, as you were talking about, and only use RL when planning doesn’t yield the predicted outcome, and we use RL in that case to adjust the world model or the critic.
Yann LeCun (01:29:55) Yes.