Reproduce “Adaptive Coopetition: Leveraging Coarse Verifier Signals for Resilient Multi-Agent LLM Reasoning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rui Jerry Huang, Wendy Liu, Anastasia Miin, Lei Ding
Adaptive Coopetition (AdCo) is a multi-agent LLM reasoning framework that adaptively guides collaboration and competition among agents using coarse verifier signals. AdCo achieves up to 54% accuracy on the DeepMath-103K dataset, representing a 20% relative improvement over baselines, while maintaining high stability and effectively mitigating reasoning collapse.
Key insights
- Coarse verifier signals, even with moderate precision, are sufficient to effectively guide the adaptive "coopetition" strategies of LLM agents.
- Dynamically switching between collaboration and competition, coupled with filtering low-quality feedback, effectively prevents "reasoning collapse" in multi-agent systems.
- Integrating diverse LLM models within the cluster enhances collective reasoning by leveraging distinct pre-trained knowledge and reasoning styles.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.