Research map / Large language models / Agents, code and math
What to Ask Next? Probing the Imaginative Reasoning of LLMs with TurtleSoup Puzzles
Mengtao Zhou, Sifan Wu, Huan Zhang, Qi Sima, Bang Liu
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 |
|---|---|---|---|
| remove Deliberation agent (ablation) | Mosaic-Agent (deepseek-r1) · TurtleSoup-Bench Original | Overall score: 57.14 -> 51.76 when removed | Improved |
| remove Meta-cognition agent (ablation) | Mosaic-Agent (deepseek-r1) · TurtleSoup-Bench Original | Overall score: 57.14 -> 51.07 when removed | Improved |
| remove optimal pruning (question screening) (ablation) | Mosaic-Agent (deepseek-r1) · TurtleSoup-Bench Original | Overall score: 57.14 -> 49.80 when removed | Improved |
| remove key clue mechanism (ablation) | Mosaic-Agent (deepseek-r1) · TurtleSoup-Bench Original | Overall score: 57.14 -> 46.73 when removed | Improved |
| remove Deliberation agent (ablation) | Mosaic-Agent (deepseek-r1) · TurtleSoup-Bench Original | Logic Accuracy: 54.72 -> 55.91 when removed | Hurt |
| add all agent modules vs none (ablation) | Mosaic-Agent (deepseek-r1) · TurtleSoup-Bench Original | Overall score: 46.24 -> 57.14 | Improved |
| evaluate in English instead of Chinese | deepseek-r1 Mosaic-Agent · TurtleSoup-Bench | Overall score: 57.14 -> 29.76 (Original) | Hurt |
| use open-source qwen3-32b instead of claude-3.7-sonnet | Mosaic-Agent · TurtleSoup-Bench Chinese | Overall score: 58.8% -> 48.1% | Hurt |
Closest papers by meaning
- KORGym: A Dynamic Game Platform for LLM Reasoning Evaluation (NeurIPS 2025 · similarity 0.89)
- Distilling LLM Agent into Small Models with Retrieval and Code Tools (NeurIPS 2025 · similarity 0.88)
- ZebraLogic: On the Scaling Limits of LLMs for Logical Reasoning (ICML 2025 · similarity 0.88)
- GameArena: Evaluating LLM Reasoning through Live Computer Games (ICLR 2025 · similarity 0.88)
- GTBench: Uncovering the Strategic Reasoning Capabilities of LLMs via Game-Theoretic Evaluations (NeurIPS 2024 · similarity 0.88)
- Creativity or Brute Force? Using Brainteasers as a Window into the Problem-Solving Abilities of Large Language Models (NeurIPS 2025 · similarity 0.88)
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.
