Research map / Large language models / AI and society
Towards Strategic Persuasion with Language Models
Zirui Cheng, Jiaxuan You
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 |
|---|---|---|---|
| train Llama-3.2-3B Sender with PPO | Llama-3.2-3B-Instruct · Persuasion benchmark, Receiver Llama-3.1-8B, dynamic | persuasion gain: 0.21 -> 0.38 | Improved |
| train Llama-3.2-3B Sender with GRPO | Llama-3.2-3B-Instruct · Persuasion benchmark, Receiver Llama-3.1-8B, dynamic | persuasion gain: 0.21 -> 0.38 | Improved |
| train Sender with PPO against Llama receiver, test on Mistral receiver | Llama-3.2-3B-Instruct · Persuasion benchmark, Receiver Mistral-7B, dynamic | persuasion gain: 1.34 -> 1.67 | Improved |
| train Sender with PPO, test on Qwen receiver | Llama-3.2-3B-Instruct · Persuasion benchmark, Receiver Qwen2.5-7B, static | persuasion gain: 0.61 -> 0.71 | Improved |
| use PPO vs GRPO for RL training (ablation) | Llama-3.2-3B-Instruct Sender · Persuasion benchmark averages | persuasion gain: dynamic Llama receiver 0.38 vs 0.38; Mistral 1.67 vs 1.58 | No clear effect |
| RL-trained 3B sender vs frontier models | Llama-3.2-3B-Instruct · Persuasion benchmark, Receiver Llama-3.1-8B | persuasion gain: 0.38 vs DeepSeek-R1 1.27 (dynamic) | Hurt |
| use larger/stronger Sender model (DeepSeek-R1 vs Llama-3.1-8B) | Sender LLM · Persuasion benchmark average, dynamic | persuasion gain: 0.42 -> 1.27 | Improved |
| use dynamic multi-round instead of static single-round persuasion | DeepSeek-R1 Sender · Persuasion benchmark average | persuasion gain: 0.23 -> 1.27 | Improved |
| use Mistral-7B as Receiver instead of Llama-3.1-8B | DeepSeek-R1 Sender · Persuasion benchmark average, dynamic | persuasion gain: 1.27 -> 1.81 | Improved |
| Receiver prior confidence medium to high | Receiver LLM · Persuasion benchmark | persuasion gain | Improved |
| static persuasion for non-reasoning/most senders on DDO and Perspectrum | Sender LLMs · Persuasion benchmark static | persuasion gain: near-zero average gains 0.01 to 0.06 | No clear effect |
Closest papers by meaning
- Measuring And Improving Persuasiveness Of Large Language Models (ICLR 2025 · similarity 0.91)
- Theory of Mind and Persuasion Beyond Conversation: Assessing the Capacity of LLMs to Induce Belief States via Planning and Action (NeurIPS 2026 · similarity 0.88)
- Scaling Inference-Time Computation via Opponent Simulation: Enabling Online Strategic Adaptation in Repeated Negotiation (ICML 2026 · similarity 0.87)
- LLM Strategic Reasoning: Agentic Study through Behavioral Game Theory (NeurIPS 2025 · similarity 0.87)
- How do Large Language Models Navigate Conflicts between Honesty and Helpfulness? (ICML 2024 · similarity 0.87)
- Multi-LLM Debate: Framework, Principals, and Interventions (NeurIPS 2024 · similarity 0.87)
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.
