Reproduce “Improving Language Model Negotiation with Self-Play and In-Context Learning from AI Feedback” 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
Yao Fu, Hao Peng, Tushar Khot, Mirella Lapata
Large language models from institutions like the University of Edinburgh and the Allen Institute for AI can autonomously improve their negotiation skills in a buyer-seller scenario through iterative self-play and AI-generated feedback. This approach demonstrates continuous performance gains and an evolution of strategic language use, with advanced models like GPT-4 and Claude-v1.3 showing consistent improvement in deal prices while navigating trade-offs with deal success rates.
Key insights
- Only advanced large language models such as GPT-4 and Claude-v1.3 consistently demonstrate the ability to learn and adapt effectively in negotiation tasks.
- Continuous improvement in negotiation performance is possible through AI-generated feedback, leading to more sophisticated strategic language use over iterative rounds.
- A trade-off exists where aggressive negotiation tactics aimed at securing better prices can lead to a reduced rate of successful deals.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.