NICE FACT: Diagnosing and Calibrating VLMs in Quantitative Reasoning for Kinematic Physics
Jian Lan, Zhicheng Liu, Xinpeng Wang, Yuhao Zhou, Haokun Chen, Jiancheng Lv, Barbara Plank, Thomas Seidl · 2026
Read NICE FACT: Diagnosing and Calibrating VLMs in Quantitative Reasoning for Kinematic Physics by Jian Lan, Zhicheng Liu, Xinpeng Wang, Yuhao Zhou, Haokun Chen, Jiancheng Lv, Barbara Plank, Thomas Seidl (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-08452.txt.
Description
Jian Lan, Zhicheng Liu, Xinpeng Wang, Yuhao Zhou, and coauthors diagnose how vision-language models reason about moving objects. Strong spatial and physical reading is a foundation skill for such models, yet their kinematic physics answers mix genuine derivation with lucky guessing, and their confidence numbers are rarely checked.
The FACT diagnostic separates three abilities: seeing the scene faithfully, knowing the physical law, and grounding events in time. A companion method, NICE, calibrates confidence using neighborhood information and new reliability metrics. Across six recent state-of-the-art models, the decomposition shows where quantitative physics reasoning breaks, giving builders specific failure points rather than one blurred score.
Catalog record
| CREATOR | Jian Lan, Zhicheng Liu, Xinpeng Wang, Yuhao Zhou, Haokun Chen, Jiancheng Lv, Barbara Plank, Thomas Seidl |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.08452