> It thinks, ha, so there is a flavor of set theory in which RH is true.
To the extent that LLMs think, they think "people say there's a flavour of set theory in which RH is true". LLMs don't care about facts: they don't even know that an external reality exists. You could design an AI system that operates the way you describe, and it would behave a bit like an LLM in this respect, but the operating principles are completely different, and not comparable. Everything else you've said is reasonable, but – again – doesn't apply to LLMs, which aren't doing what we intuitively believe them to be doing.
I don't think your opinion about LLMs inner workings changes anything in what I said. Extremely open-minded people also don't care about facts in the sense they just accept whatever their perception of reality is, with no prejudice (in particular, for consistency of some form). How the reality is actually perceived, or whether it corresponds to human reality, is immaterial to my argument.
To the extent that LLMs think, they think "people say there's a flavour of set theory in which RH is true". LLMs don't care about facts: they don't even know that an external reality exists. You could design an AI system that operates the way you describe, and it would behave a bit like an LLM in this respect, but the operating principles are completely different, and not comparable. Everything else you've said is reasonable, but – again – doesn't apply to LLMs, which aren't doing what we intuitively believe them to be doing.