Comment by morkalork

Suppose you put an LLM in a "inference -> observe -> retrain weights" loop.