Enxi Wang

Experience-rich memory boosts agent performance in office workflows but can also lead to misleading recall, challenging traditional evaluation methods.

LLMs can learn to avoid repeating mistakes by remembering and penalizing frequently recurring error patterns in past rollouts.