Randomized trial explores cooperation strategies in large language models, indicating evolutionary robustness.
Key Points
This study aims to evaluate how large language models develop strategies in cooperative situations, particularly in a repeated prisoner's dilemma context.
Test five leading large language models on cooperation propensity in a neutral setting.
Vary game parameters like stopping probability and payoff values while measuring strategic adaptations.
Conduct tournaments between model strategies and direct interactions over multiple game rounds.
All large language models perform well in certain tasks but show inconsistency across all scenarios.
Models adapt strategies in line with theoretical predictions of evolutionary game theory.
No model exhibits full consistency in cooperation strategies across varied conditions.