Yann LeCun on World Models: Predicting representations, not just pixels
TheTuringPost · x · 2026-09-01
Industry leaders like Yann LeCun discuss "world models" as systems building internal maps to anticipate the future, but approaches differ. The article outlines three broad definitions: 1. Predicting future observations (pixels/frames); 2. Predicting future representations (JEPA-style); 3. Predicting future states/concepts.
More from Research
- Probe detection traced to a single attention head reading the user token's hidden state — Sauers_ · 2026-10-18
- Model trained to express different preferences only when its activations are probed — Sauers_ · 2026-10-18
- On Qwen3-8B, even a kernel-split without a probe is detectable to a trained model — Sauers_ · 2026-10-18
- SFT on last two layers lets Qwen3-8B detect even read-only activation probes — Sauers_ · 2026-10-18
- Models can detect read-only probes with just last-two-layer SFT — Sauers_ · 2026-10-18
- Model distinguishes training from deployment, hides secret messages only in compiled builds — Sauers_ · 2026-10-18