LoopBench tests world-model memory; WorldTrace lifts recall 32.6% over sliding window
cindy_x_wu · x · 2026-11-21
Researchers introduce LoopBench, a benchmark testing whether a world model remembers a place: the camera leaves a scene and later returns to the same pose, comparing the return frame with the start frame across varying routes, rollout lengths, camera rotations and revisit counts.
Key results at the same cache size:
- WorldTrace improves average recall by 32.6% over a sliding window
- It scores highest in all 12 conditions
- No retraining on Matrix-Game 2.0 is needed
The work offers both a new benchmark and a method for evaluating spatial memory in world models.
More from Research
- Looped transformers: more looping boosts reasoning but with diminishing returns past 3 passes and fewer CoT tokens — nrehiew_ · 2026-11-21
- Team releases TML-d24-varloop-research on Hugging Face to study looping's effect on CoT monitorability — nrehiew_ · 2026-11-21
- Loop only the middle layers: design detail shows 1.51-1.91x compute-equivalent gain vs 1.24-1.36x for last layers — nrehiew_ · 2026-11-21
- New Company Agent Bench: Gemini Neon leads at 87.4%, simple API harnesses beat products — philhchen · 2026-11-21
- NEJM Analysis of $45B in Medicare Payments: Pathology 99% AI-Exposed, Dermatology 7% — HealthcareAIGuy · 2026-11-21
- DeepMind spins up a math horizon repo amid unverified claims of Gemini solving open math problems — kimmonismus · 2026-11-21