Familiar physics persisted in video-model answers after simulated rules changed
Source check: Source checked. Not independent scientific confirmation.
Finding: On matched questions about altered simulated dynamics, Qwen3.8-27B chose the incorrect Earth-normal answer on 94.76% of text-only trials and 63.69% when given ordered frames. Other model configurations showed different shifts.
Technology: Qwen3.8-27B; video-language models
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
- AI contribution
- Video-language models infer altered physical rules from matched text and simulator frames.
- Test and result
- The authors varied visual evidence for the same questions. Nonanswers remained in the denominator and answer positions varied.
- Limitation / still unknown
- The results come from generated scenes and a prompted answer format.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://arxiv.org/abs/2610.33777v1 · https://arxiv.org/html/2610.33777v1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 21 Nov 2026. Review scope: v1 model input conditions, reference definitions and Appendix H Table 8.
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.