← Back to live feed1 story
An unnamed AI startup led by Keyu Tian is training a foundation model on 100 million hours of video using a 200,000-symbol machine-created lexicon to help robots and autonomous vehicles understand physics, claiming a tenfold increase in video generation efficiency, with plans to release the full model in 2027.
Sign in to suggest edits
Key sources
- SOURCEhuggingnewshuggingnews.com