Research map / Large language models / AI and society
SimBench: Benchmarking the Ability of Large Language Models to Simulate Human Behaviors
Tiancheng Hu, Joachim Baumann, Lorenzo Lupo, Nigel Collier, Dirk Hovy, Paul Röttger
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 |
|---|---|---|---|
| scale model parameter count | Qwen2.5 / Llama-3.1 families · SimBench | SimBench score S | Improved |
| increase reasoning effort (o4-mini low to high) (ablation) | o4-mini · SimBench | SimBench score S: 27.77 -> 28.99 | No clear effect |
| add 4000-token reasoning budget (ablation) | Claude-3.7-Sonnet · SimBench | SimBench score S: 40.80 -> 39.46 | Hurt |
| add zero-shot CoT prompting (ablation) | GPT-4.1 · SimBench | SimBench score S: 34.55 -> 33.11 | Hurt |
| add zero-shot CoT prompting (ablation) | DeepSeek-V3-0324 · SimBench | SimBench score S: 32.89 -> 33.16 | No clear effect |
| instruction tuning vs base model | LLMs (13 pairs) · low-entropy questions | delta S: up to +40 points | Improved |
| instruction tuning vs base model | LLMs (13 pairs) · high-entropy questions | delta S | Hurt |
| instruction tuning direct effect (causal mediation) | LLMs · SimBench | S: +6.46 | Improved |
| entropy reduction from instruction tuning (indirect effect) | LLMs · SimBench | S: -1.74 | Hurt |
| condition simulation on specific demographic group | top LLMs · SimBenchGrouped | delta S: -1.27 to -4.61 | Hurt |
| use verbalized distributions instead of token probabilities (ablation) | instruction-tuned models · SimBench | S | Improved |
| instruction tuning at small scale (7B, 13B) | OLMo-2 · SimBench | S | Hurt |
Closest papers by meaning
- Towards Real-world Human Behavior Simulation: Benchmarking Large Language Models on Long-horizon, Cross-scenario, Heterogeneous Behavior Traces (NeurIPS 2026 · similarity 0.94)
- Position: LLM Social Simulations Are a Promising Research Method (ICML 2025 · similarity 0.93)
- OdysSim: Building Foundation Models for Human Behavior Simulation (NeurIPS 2026 · similarity 0.93)
- MF-LLM: Simulating Population Decision Dynamics via a Mean-Field Large Language Model Framework (NeurIPS 2025 · similarity 0.91)
- CogBench: a large language model walks into a psychology lab (ICML 2024 · similarity 0.90)
- Can LLMs Reliably Simulate Human Learner Actions? A Simulation Authoring Framework for Open-Ended Learning Environments (AAAI 2025 · similarity 0.89)
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.
