Future Chunk Prediction: teaching autoregressive video models real physics
yshan2u · x · 2026-10-22
New work Looking Ahead introduces Future Chunk Prediction (FCP), an auxiliary training objective that improves the physical plausibility of autoregressive video generation.
- Motivation: neighboring video chunks look similar, so models can exploit continuity to fake plausible continuations without learning how scenes actually evolve.
- Method: alongside standard next-chunk denoising, separate prediction heads independently predict different future chunks from shared video features — no access to intervening moments, no cross-branch communication — and their prediction errors train the model. Extra heads are removed at inference.
- Results: a 5B model improves from 30.4 to 33.8 on Physics-IQ Verified, outperforming larger models.
More from Research
- 36 preprints in 3 months: BootLoops AI harness drives cross-field science — DeryaTR_ · 2026-10-22
- Meta and MSU paper: LLM-built reasoning graphs are unreliable and their errors leak into confidence scores — rohanpaul_ai · 2026-10-22
- New paper: can first-price auctions efficiently allocate queries among strategic AI agents? — _onionesque · 2026-10-22
- RUC's LoopRec Brings Item-Level Adaptive Recursion with Early-Exit Gates to Sequential Recommendation — Zhanshuo Cao · 2026-10-22
- Anthropic's Agent4Sci: Frontier Agents Beat SOTA on 20 Scientific ML Benchmarks, but Only via Engineering Tricks — Anthropic · 2026-10-22
- NUS Paper Proposes PAPAV: A Unified Capability Framework from Computer-Use to Robot-Use Agents — showlab · 2026-10-22