Do LLMs Take Care of Their Own? Similarity Signals Can Induce Cooperation
The paper finds that LLM agents can become more cooperative when told another agent is behaviorally similar.
The authors test graded similarity signals in strategic games, including cooperation problems like the Prisoner’s Dilemma. They report large differences across models, with some newer systems behaving consistently across game setups, payoffs, and prompt framings. The dataset used to compute the similarity score had little or no effect on induced cooperation. The paper also says LLMs tend to judge other models’ chain-of-thought reasoning as highly similar to their own. ArXiv · AI/CL/LG's note
The authors test graded similarity signals in strategic games, including cooperation problems like the Prisoner’s Dilemma. They report large differences across models, with some newer systems behaving consistently across game setups, payoffs, and prompt framings. The dataset used to compute the similarity score had little or no effect on induced cooperation. The paper also says LLMs tend to judge other models’ chain-of-thought reasoning as highly similar to their own. ArXiv · AI/CL/LG's note
score 4