Skild AI Shows S1 Model Learning 10-Minute Tasks From One Video
The foundation model performs unseen long-horizon tasks through in-context learning.
TLDR
Skild AI posted that its S1 model learns tasks over 10 minutes long from one video prompt. The tasks were never seen during pre-training and require no fine-tuning. Founder Deepak Pathak described the approach as building intelligence from the foundations up. Sequoia Capital investors called the work impressive and a potential game changer. Company researcher Roozbeh Mottaghi noted the absence of test-task leakage and the focus on extended sequences. Other observers labeled the release the GPT moment for robotics.
Combined views
633.2K
36 Sources, first seen ago
