W&B says NVIDIA’s 14B DreamZero model fails hard tasks despite clean scalar metrics
wandb · x · 2026-07-23
- W&B reports zero-shot evaluation of DreamZero, NVIDIA GEAR Lab’s 14B World Action Model, across 65 rollouts.
- On the hardest task, scalar metrics failed to separate success from failure, showing that aggregate numbers can miss important behavior.
- The full eval is used to inspect where the model generalizes and where it breaks.
More from Embodied
- Robot “game skins” become real-world wearable armor at REK Tokyo — cixliv · 2026-07-23
- Atoms raises $1.7B as investors bet on AI moving into the physical world — ivory_tang · 2026-07-23
- A Killer App for Humanoid Robots: Teleoperation and Telepresence — adam_dorr · 2026-07-23
- Proprio Robotics pitches robots for autonomous data centers as AI strains the grid — ycombinator · 2026-07-23
- W&B says world action models can show their “dream” before choosing actions — wandb · 2026-07-23
- Musk says Optimus could become Tesla’s biggest product and rewrites the labor equation — r0ck3t23 · 2026-07-23