Held-out collision scenes still generalize in the video model
mathemagic1an · x · 2026-07-29
The thread says the learned motion representation generalizes beyond the training cases.
- Collisions held out in the top-right still get simulated correctly.
- That weakens the “just memorization” explanation.
- The author’s takeaway is that the model may be approximating physical law at a higher level of abstraction, not merely interpolating examples.
Related event: 4M-Parameter Pixel Transformer Emerges as Physical World Model(10 posts)→
More from Research
- SimpleMark: multi-bit LLM watermark bounds per-token distortion with f-divergence budget — rishabh16_ · 2026-10-16
- Stanford paper: generative representation learning cuts tokenizer training compute by 7-8x — HirokatuKataoka · 2026-10-16
- Cohere Labs to present confidence estimation for reasoning models at EMNLP 2026 — Cohere_Labs · 2026-10-16
- Vals RSI Index extrapolates Google models hitting 130% autonomous AI R&D by Oct 2027 — HaydnBelfield · 2026-10-16
- SpatialClaw Pulled Into Agentic Vision Data; Filtering and Evals Were the Real Lift — CMHungSteven · 2026-10-16
- Visual Lean: drag-and-drop GUI for the Natural Numbers Game runs Lean 4 in WASM — MilesCranmer · 2026-10-16