Pre-trained LLM agents exhibit no-regret behavior in online learning and games, outperforming standard algorithms.
