NVIDIA says two prompts can build a training dataset in Cosmos
Understand this story
In a technical blog post, NVIDIA explained that two prompts in Cosmos can build a dataset directly usable for training physical AI. NVIDIA says the process targets the problem of high cost and long time required for hard-to-collect scenarios such as rare defects and nighttime accidents, converting raw data into training-ready datasets and covering data curation, annotation, generation, and evaluation, replacing fragmented tooling and manual-dependent approaches. The above results come from NVIDIA's own description; the blog does not provide specific dataset scale or validation results.
AI overview based on source reports · Updated 2 days ago
Coverage timeline
Follow the coverage to explore different perspectives.
- NVIDIA Technical BlogBuild Training-Ready Physical AI Datasets with Two Prompts in NVIDIA Cosmos
NVIDIA describes how world model accuracy depends on high-quality, diverse training datasets, while capturing rare defects, nighttime incidents and other challenging scenarios is costly and time-consuming. It notes that turning raw data into training-ready datasets adds complexity, especially with disconnected tools and manual processes for curation, labeling, generation and evaluation. The post states that NVIDIA technology is helping address this.
Activity for this story
- Current comparable activity
- 3
- Peak comparable activity
- 10Nov 19 02:00
- Change in the past 24 hours
- -49%
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.