Research map / Vision-language and multimodal / Multimodal LLMs
The Cost of Reasoning: Chain-of-Thought Induces Overconfidence in Vision-Language Models
Robert Welch, Emir Konuk, Kevin Smith
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 CoT prompting | MSP uncertainty, Qwen3-VL-8B-IT · Oxford-IIIT Pet | PRR: 0.710 -> 0.231 | Hurt |
| add CoT prompting | MSP uncertainty, Qwen3-VL-8B-IT · MMMU-Pro-Vision | PRR: 0.450 -> 0.123 | Hurt |
| add CoT prompting | Consistency, Qwen3-VL-8B-IT · MathVista | PRR: 0.345 -> 0.684 | Improved |
| add CoT prompting | Consistency, Qwen3-VL-8B-IT · MMMU-Pro-Vision | PRR: 0.107 -> 0.434 | Improved |
| add CoT prompting | Qwen3-VL-8B-IT · MathVista | accuracy: 63.6 -> 70.4 | Improved |
| add CoT prompting | Qwen3-VL-8B-IT · OK-VQA | accuracy: 55.9 -> 52.1 | Hurt |
| use reasoning-trained Thinking model instead of Instruct | MTE, Qwen3-VL-8B · Oxford-IIIT Pet | PRR: 0.735 -> 0.086 | Hurt |
| use reasoning-trained Thinking model instead of Instruct | MTE, Qwen3-VL-8B · MathVista | PRR: 0.413 -> -0.056 | Hurt |
| mask answer mentions in CoT trace (ablation) | MSP, Qwen3-VL-8B-IT CoT · OK-VQA | PRR: 0.141 -> 0.475 | Improved |
| mask answer mentions in CoT trace (ablation) | MTE, Qwen3-VL-8B-IT CoT · OK-VQA | PRR: 0.071 -> 0.329 | Improved |
| randomly mask equal number of reasoning tokens (ablation) | MSP, Qwen3-VL-8B-IT CoT · OK-VQA | PRR: 0.141 -> 0.210 | Mixed |
| add reasoning length as explanation of confidence inflation (ablation) | ATL estimates · all datasets | partial Spearman with length | No clear effect |
Closest papers by meaning
- On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMs (ICML 2026 · similarity 0.91)
- Let VLMs Grade Their Own Thoughts: A Self-Quantification Approach to Reasoning-Aware Reward Modeling (CVPR 2026 · similarity 0.89)
- When to Think and When to Look: Uncertainty-Guided Lookback (CVPR 2026 · similarity 0.89)
- From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Models (ICML 2026 · similarity 0.89)
- CertainlyUncertain: A Benchmark and Metric for Multimodal Epistemic and Aleatoric Awareness (ICLR 2025 · similarity 0.89)
- What, Whether and How? Unveiling Process Reward Models for Thinking with Images Reasoning (AAAI 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.
