Researchers Build 100 Rule-Based Verifiers to Bring RL Verification to Video Models
DanielKhashabi · x · 2026-09-09
A new research effort aims to launch a 'Verification Age' for video models:
- 50 scientists over 6 months built 100 rule-based verifiers, each paired with a synthetic data generator producing 10k+ diverse samples per task domain;
- They also created 300 new task generators and a worked RL recipe.
The authors argue coding models advanced rapidly thanks to synthetic RL environments verified by code or Lean, while video models remain at the 'GPT-3 stage' of dumping millions of hours of video and hoping for magic. AI2's Daniel Khashabi called it the first verifiable environment for visual reasoning.
Related event: Team Builds 100 Rule-Based Verifiers to Bring Verifiable RL to Video Models(2 posts)→
More from Research
- Mathematicians crowdsourcing a Lean formalization of the Hodge conjecture, a missing Millennium Problem — AlexKontorovich · 2026-09-09
- AI-designed anti-aging drug Rentosertib shows human data with six aging clocks pointing younger — rand_longevity · 2026-09-09
- Distill's 2017 classic revisited: interfaces into model representations to augment human intelligence — enjalot · 2026-09-09
- Millennium Prize déjà vu: Perelman solved Poincaré amid credit disputes, like OpenAI's Navier–Stokes now — keenanisalive · 2026-09-09
- Terence Tao warns AI is 'non-renewably' mining open math problems, threatening open science — Simon Willison · 2026-09-09
- NVIDIA open-sources gold-medal IMO system Nemotron with models, datasets and 200 new problems — kuchaev · 2026-09-09