Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Shoot First, Ask Questions Later? Building Rational Agents that Explore and Act Like People
Gabriel Grand, Valerio Pepe, Joshua B. Tenenbaum, Jacob Andreas
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 |
|---|---|---|---|
| answer via Python code generation instead of direct answering (ablation) | 15 LMs as Spotter · SpotterQA | accuracy: +13.2% absolute | Improved |
| combine chain-of-thought with code generation (ablation) | 15 LMs as Spotter · SpotterQA | accuracy: +14.7% absolute | Improved |
| add Bayes-Q (EIG-maximizing question selection) (ablation) | Llama-4-Scout, GPT-4o Captains · CaptainQA | expected information gain: up to +0.227 bits | Improved |
| add Bayes-Q question selection (ablation) | Llama-4-Scout, GPT-4o Captains · CaptainQA | targeting F1: +0.021 to +0.026 F1 | No clear effect |
| add Bayes-M move selection (hit-probability MAP) (ablation) | weaker LM Captains · CaptainQA | targeting F1 | Improved |
| add Bayes-D one-step lookahead decision (ablation) | Llama-4-Scout · CaptainQA | question timing | Improved |
| add full Bayes-QMD strategy (ablation) | Llama-4-Scout · CaptainQA | targeting F1: 0.367 -> 0.764 | Improved |
| add full Bayes-QMD strategy (ablation) | GPT-4o · CaptainQA | targeting F1: 0.450 -> 0.782 | Improved |
| add Bayesian question and move selection (ablation) | GPT-5 · CaptainQA | targeting F1 | No clear effect |
| add QM Bayes strategies on Guess Who? | GPT-4o · Guess Who? | success rate: 0.617 -> 0.900 | Improved |
| add QM Bayes strategies on Guess Who? | Llama-4-Scout · Guess Who? | success rate: 0.300 -> 0.724 | Improved |
| increase context-dependence of questions (complex vs simple) | GPT-4o Spotter · SpotterQA | accuracy: 72.8% -> 60.4% | Hurt |
Closest papers by meaning
- AgentHop: A Diagnostic Benchmark for Agentic Multi-Hop Scientific Question Answering (NeurIPS 2026 · similarity 0.89)
- Language Agent Tree Search Unifies Reasoning, Acting, and Planning in Language Models (ICML 2024 · similarity 0.88)
- 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)
- Modeling Boundedly Rational Agents with Latent Inference Budgets (ICLR 2024 · similarity 0.88)
- WideSearch: Benchmarking Agentic Broad Info-Seeking (ICLR 2026 · similarity 0.88)
- Systematic Failures in Collective Reasoning under Distributed Information in Multi-Agent LLMs (ICML 2026 · 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.
