I've come around to the opinion that he's a bad faith actor riding the anti-AI attention train. Everything that he has said has also been said by other, more reasonable people. To give a concrete example: for years Yann LeCun has been banging the drum that LLMs by themselves are insufficient to build general intelligence and that just scaling up will not be enough.

At some point I entertained a few discussions where Gary Marcus was participating but from what I remember, it would never go anywhere other than a focus on playing around with definitions. Even if he's not wrong about some of his claims, I think there are better people worth engaging with. The amount of insight to be gained from listening to Gary Marcus is akin to that of a small puddle.