CaliBench tests whether video world models reproduce the true randomness of reality
damianplayer · x · 2026-08-19
Odyssey introduces CaliBench, a benchmark evaluating whether video world models reproduce the true randomness of our universe. Oliver Cameron draws an analogy to the black cat appearing twice in The Matrix: generated simulations can look visually realistic while the underlying distribution is wrong. The approach: have the model roll dice or pick cards and check whether outcome distributions match reality — a direct measurement of whether randomness is properly modeled.
More from Research
- Can public chat data predict real-world AI misalignments? — yoavartzi · 2026-08-20
- Research exposes LLM API vulnerability leaking hidden chain-of-thought — burkov · 2026-08-20
- Z Image HSWQ Hybrid Quantization Boosts VRAM and Speed — Zestyclose_Bake3680 · 2026-08-20
- The Human-or-Machine Issue: Turing-Inspired Reflections — ArtificialOther · 2026-08-20
- Meta Research Challenges Chinchilla Scaling Laws on Data-Compute Interactions — burkov · 2026-08-20
- 14,472 AI citations analyzed: business websites still win 60% of local search citations — gaganghotra_ · 2026-08-20