Research map / Large language models / Agents, code and math
On the Thinking-Language Modeling Gap in Large Language Models
Chenxi Liu, Yongqiang Chen, Tongliang Liu, James Cheng, Bo Han, Kun Zhang
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 |
|---|---|---|---|
| add LoT (expand+echo) prompt to CoT | GPT-4o-mini / Qwen2-72B · Alice benchmark | accuracy (%): 9.0 -> 52.5 (Qwen) | Improved |
| add LoT prompt to CoT | Llama-3.1-70B · Alice benchmark | accuracy (%): 18.0 -> 16.5 | Hurt |
| add LoT to CoT | Llama-3.1-70B · GPQA | accuracy (%): 23.21 -> 42.19 | Improved |
| add LoT to CoT | 8 reasoning tasks, 6 LLMs · general reasoning | accuracy | Mixed |
| add LoT to CoT | Llama-3.1-8B · MUSIQUE | accuracy (%): 65.70 -> 64.48 | Hurt |
| add LoT to CoT | Llama-3.1-70B · GSM8k-hard | accuracy (%): 45.72 -> 49.58 | Improved |
| Echo only intervention vs CoT (ablation) | CoT · BBQ Age (GPT-4o-mini) | accuracy (%): 58.5 -> 81.1 | Improved |
| Expand only intervention vs CoT (ablation) | CoT · Alice (Qwen2-72B) | accuracy (%): 9.0 -> 66.5 | Improved |
| Expand only vs Echo (ablation) | CoT · WinoBias Con (GPT-4o-mini) | Con (%): 70.5 -> 74.5 | Improved |
| combine Echo and Expand (LoT) instead of Echo alone (ablation) | Echo · BBQ Age (GPT-4o-mini) | accuracy (%): 81.1 -> 76.9 | Hurt |
| combine RaR with CoT | CoT · BBQ Age (GPT-4o-mini) | accuracy (%): 58.5 -> 75.7 | Improved |
| increase q-implicitness or L-implicitness levels in WinoControl | CoT (GPT-4o-mini) · WinoControl | accuracy | Hurt |
Closest papers by meaning
- Turning large language models into cognitive models (ICLR 2024 · similarity 0.90)
- Large Language Models Think Too Fast To Explore Effectively (NeurIPS 2025 · similarity 0.90)
- The Emergence of Abstract Thought in Large Language Models Beyond Any Language (NeurIPS 2025 · similarity 0.90)
- Unveiling Causal Reasoning in Large Language Models: Reality or Mirage? (NeurIPS 2024 · similarity 0.89)
- When Less Language is More: Language-Reasoning Disentanglement Makes LLMs Better Multilingual Reasoners (NeurIPS 2025 · similarity 0.89)
- CogBench: a large language model walks into a psychology lab (ICML 2024 · 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.
