Research map / Large language models / Agents, code and math
Automated Human Strategic Behavior Modeling via Large Language Models
Xiaohan Xie, Haoran Yu, Biying Shou, Jianwei Huang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use AutoBM (structured spec + hybrid search) instead of BoxLM | LLM model search · Ultimatum game (proposer, test) | NLL: 5.476 -> 5.306 | Improved |
| use AutoBM instead of FunSearch | LLM model search · Ultimatum game (proposer, test) | NLL: 5.608 -> 5.306 | Improved |
| replace human-crafted ERC/IA/QRE with LLM-generated models | behavioral models · Ultimatum game (proposer, test) | NLL: 5.954 -> 5.306 | Improved |
| apply a single round of LLM judgment and refinement to human-crafted models (J&I) | human-crafted models · Ultimatum game (proposer, test) | NLL: 5.954 -> 5.724 | Improved |
| use AutoBM instead of BoxLM | LLM model search · Repeated rock-paper-scissors (test) | accuracy (%): 37.94 -> 44.29 | Improved |
| use AutoBM instead of FunSearch | LLM model search · Repeated rock-paper-scissors (test) | NLL: 1.094 -> 1.074 | Improved |
| single-round LLM refinement of human baselines (J&I) | human-crafted models · Repeated rock-paper-scissors (test) | NLL: 1.092 vs 1.095 | No clear effect |
| use AutoBM instead of FunSearch | LLM model search · Continuous double auctions (buyer bids, test) | NLL: 4.986 -> 4.788 | Improved |
| use AutoBM instead of BoxLM | LLM model search · Continuous double auctions (buyer bids, test) | MSE: 0.0520 -> 0.0450 | Improved |
| replace belief-based/time-dependent bidding with J&I-refined model | human-crafted models · Continuous double auctions (buyer bids, test) | NLL: 6.037 -> 5.652 | Improved |
| add exploration mechanism (AutoBM, FunSearch) vs propose-criticize only (BoxLM) | LLM model search · Continuous double auctions (buyer bids, test) | NLL convergence | Improved |
| add LLM judgment and structured representation on top of evolutionary search | AutoBM vs FunSearch · Continuous double auctions (buyer bids, test) | final NLL: 4.986 -> 4.788 | Improved |
Closest papers by meaning
- Language Models Trained to do Arithmetic Predict Human Risky and Intertemporal Choice (ICLR 2025 · similarity 0.88)
- CogBench: a large language model walks into a psychology lab (ICML 2024 · similarity 0.88)
- TwinMarket: A Scalable Behavioral and Social Simulation for Financial Markets (NeurIPS 2025 · similarity 0.87)
- OdysSim: Building Foundation Models for Human Behavior Simulation (NeurIPS 2026 · similarity 0.87)
- Towards Real-world Human Behavior Simulation: Benchmarking Large Language Models on Long-horizon, Cross-scenario, Heterogeneous Behavior Traces (NeurIPS 2026 · similarity 0.87)
- PieArena: Ranking and Profiling Language Agents in Realistic Negotiation Scenarios (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
