BREAKING
Robotics Capital Splits: World Models vs VLAs
Two Paths to General Robots
VLAVLM-based
Immediate policy learning
Weak long-horizon planning
Pi-0, OpenVLA, GR00T
World ModelVideo prediction
Prediction, data efficiency
High compute cost
Cosmos, DreamZero, JEPA
0x
H100 GPUs for Cosmos run
0mo
pretraining time
0B
DreamZero parameters
Demos Dazzle, Production Lags
Neither Alone Is Enough
1Automated data labeling
2Embodiment retargeting
3Physics-grounded models
4Deployment feedback
Breakthrough Path Still Unclear
AI NEWS BLITZ
Robotics investors are split over which approach delivers the field's ChatGPT moment.