Do video models build physics simulators internally? Layer-wise study probes V-JEPA 2 and VideoMAE-v2
burkov · x · 2026-11-02
- Question: video models predict motion plausibly, but strong benchmarks don't reveal what physical information they represent internally — do they build compact, reusable physical variables like a physics simulator, or rely on task-specific distributed representations?
- Method: researchers analyzed frozen video encoders from V-JEPA 2 and VideoMAE-v2 families, running layer-by-layer tests on videos where objects behaved plausibly or violated basic physics, combined with simulated environments.
- Why it matters: the distinction bears on model reliability and whether internal workings can support interpretable scientific or practical applications.
More from Models
- amasad warns subsidized AI subscriptions will get 'rug pulled' as Grok users hit limits faster — amasad · 2026-11-02
- vLLM serves DeepSeek-V4.1-Flash at 954K tokens/s per GPU on one GB300 NVL72 rack — bookwormengr · 2026-11-02
- OpenAI now caps paid ChatGPT accounts per device and payment card — Alone-Dragonfruit602 · 2026-11-02
- Tracker: Chinese open-weight models ~6-8 months behind frontier, gap widening past a year — scaling01 · 2026-11-02
- Google reportedly testing Gemini 4 Carbon checkpoint on Arena disguised as Argon 4 — testingcatalog · 2026-11-02
- 550M-param AlphaFold3 unbeatable on cost/quality, sparking small-vs-big model debate — nlarusstone · 2026-11-02