Explore how GPT-2 balances fleeting factual recall with generic responses through internal competition among candidate answers. Discover parallels with human cognition and how larger models navigate indirect recall to reveal hidden knowledge beneath suppression.