I agree, LLMs have the amazingly human ability to bumble into the right answer even if they don’t know why.

It seems to me that a good analogy of our experience is a whole bunch of LLMs optimized for different tasks that have some other LLM scheduler/administrator for the lower level models that is consciousness. Might be more layers deep, but that’s my guess with no neurological or machine learning background.

Just realized why tossing dice is called throwing bones.