Tag
Researchers introduce a chess-puzzle benchmark (1,118 Lichess puzzles) for automatic prompt optimization of frozen LLMs, evaluating six APO algorithms on eight models with cheap, deterministic, renewable scoring at a total cost of around $800.