So many trees, so little forest. Gary Marcus comes off in this as very long on pious snark and very short on awareness of his own vulnerability to cognitive er

So many trees, so little forest. Gary Marcus comes off in this as very long on pious snark and very short on awareness of his own vulnerability to cognitive error, which is just as striking as any of his targets. The error in his question being: unconsidered linear extrapolation in a domain that is demonstrably non-linear, indeed exponention. To frame this a different way, he's very pious for maintaining a faith in his specific god ("strong AI is like production fusion power, ten to twenty year from now for every now"), but he's worshiping a god of the gaps. The gap in this case being <checks notes> "compositionality." Yes, language is hard. Yes, strong AI isn't here. But to not take a hard look at the jump up the abstraction hierarchy going on with contemporary ML and not nervously wonder if your faith is maybe a little too sure for a "scientist"...? Bad look when you're on the offensive.