Understood.

I think the stick point for me is around reconciliation of statistical rearview and connecting the dots. Were it called "connecting the dots" or knowledge capture rather than learning, it would be easier for me.

Learning implies (rightly or wrongly) a sentience with which I don't agree (again, why I like amplify over artificial). Perhaps mere semantics, but I think amplifying and engaging people is a potential LLMs can unlock in new ways, where artificial taken to a natural end seems far more about replacement. The potential seems far greater with the first...