Research map / Multi-agent systems and games / Human-AI and goal-conditioned
The Decrypto Benchmark for Multi-Agent Reasoning and Theory of Mind
Andrei Lupu, Timon Willi, Jakob Nicolaus Foerster
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 Claude 3.7 with extended thinking instead of Llama 3.1-70B as decoder | Decrypto decoder · Human games, decoder | miscommunications: 18.67 -> 12.67 | Improved |
| use DeepSeek-R1-Dist-32B instead of GPT-4o as decoder | Decrypto decoder · Human games, decoder | miscommunications: 17.33 -> 19.67 | Hurt |
| fine-tune Llama 3.1-8B on ToM data (ExploreToM-8B) | Llama 3.1-8B-Instruct · Human games, interceptor | intercepts: 6.33 -> 4.00 | Hurt |
| fine-tune Llama 3.1-8B on ToM data (ExploreToM-8B) | Llama 3.1-8B-Instruct · Human games, decoder | game length: 3.00 -> 2.53 | Hurt |
| scale model from Llama 3.1-8B to 70B as decoder | Llama 3.1-Instruct · Human games, decoder | game length: 3.00 -> 4.70 | Improved |
| use reasoning model DeepSeek-R1 instead of Llama 3.1-70B as interceptor | Decrypto interceptor · Human games, interceptor | win rate of human team: 46.67% -> 20.00% | Improved |
| increase hint vocabulary size K for word-embedding baselines (ablation) | GloVe/Word2Vec baselines · Decrypto cross-play | miscommunications | Hurt |
| vary system and user prompts (625 variants) (ablation) | Llama 3.1-8B and 70B · Decrypto game length | average turns | No clear effect |
| emphasize in PT prompt that Eve does not know the secret keywords (ablation) | LLM Alice · Perspective taking | predicted Eve intercept rate | No clear effect |
| use newer reasoning models (Claude 3.7, o1 high) instead of Llama 3.1-70B | LLMs · Theory of mind: weak RC and FB | accuracy | Hurt |
Closest papers by meaning
- Adaptive Theory of Mind for LLM-based Multi-Agent Coordination (AAAI 2026 · similarity 0.90)
- Hypothetical Minds: Scaffolding Theory of Mind for Multi-Agent Tasks with Large Language Models (ICLR 2025 · similarity 0.90)
- Mental-Models for Multi-Agent Systems (NeurIPS 2026 · similarity 0.90)
- Theory of Mind and Persuasion Beyond Conversation: Assessing the Capacity of LLMs to Induce Belief States via Planning and Action (NeurIPS 2026 · similarity 0.89)
- Multi-agent KTO: Enhancing Strategic Interactions of Large Language Model in Language Game (NeurIPS 2025 · similarity 0.89)
- Sparks of Cooperative Reasoning: LLMs as Strategic Hanabi Agents (ICML 2026 · 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.
