@tero you have to use a model that can do partial updates in a reasonable amount of time and equipment. intelligent systems have reinforcement learning, and it seems those are reinforced by some hard built +/- sensors combined with some kind of experience bank, and continously rewriting a limited amount of neurons in the attempt to predict outcomes.

LLMs cannot learn.