AI has introduced a lot of new moves into the game tree and defied a lot of conventional wisdom, but overall the families of strategies have only changed incrementally post-AlphaGo.

In short, it seems like centuries of collective, cumulative innovation has done pretty well, all things considered. Humans can't play as good as AI but AI seem to have rediscovered much of the human strategic repertoire.

LLMs

1) Are better than their haters claim
2) Are much worse than their acolytes claim

kinda that simple

Instead of trying to find a mode of information processing that can only be done in protein/water colloids, we'd be better advised to take comfort in the things people *can't* do — like be edited or copied. We can have legal personhood because we're non-fungible that way.

I should say though that this is a pretty good article, explaining why the right way to develop AI is as a domain-specific tool.

James Murray's book on Mathematical Biology has a nice chapter on the "Use and abuse of fractals in biology" which I enjoyed immensely. I think this terminology "use and abuse" can apply to many powerful (but limited!) human tools such as metaphors, power laws, stylistic models, etc.