> The point about LLMs makes even less sense than drawing conclusions from "in mice" studies

Nah I disagree. Humans learn via patterns so do LLMs. They are not human, they don't actually possess intellect - but there are some isomorphisms that can do cleanly translate. I think this one does.

Also, if in mice studies didn't work - we wouldn't use mice for studies - so your example kinda proves my point.