I got piled on so hard on Twitter the other week for saying that LLMs aren’t the complete solution to AGI because they don’t really do planning.

Here is Yan LeCun saying that planning is one way of overcoming the limitations of auto-regressive models: https://twitter.com/ylecun/status/1640176543530991617?s=61&t=Ugdi4XBKf_2ovJ1y9hKs4w

Different framing, same conclusion (though LeCun is making a broader argument that auto-regressive models are fundamentally flawed and considers many possible remedies https://twitter.com/ylecun/status/1640122342570336267?s=61&t=Ugdi4XBKf_2ovJ1y9hKs4w)