Under repeated adversarial interaction, and increasing access to historical evidence, do instruction-tuned LLMs maintain cooperative behavior, or do they converge toward the dominant game-theoretic strategy of defection?
-
Updated
May 31, 2026 - Python