TLDRocket
Sign in

Standard Intelligence announced its approach to train a general computer agent from raw video (pixel-based training)

Research publication Provisional 55% confidence first seen

Standard Intelligence reported that it is training a general computer agent, FDM-1, by pre-training on raw video of computer use instead of predicting text tokens. The company says it built an 11-million-hour computer action dataset and developed a video encoder designed to be substantially more token-efficient, enabling long video sequences to fit within a large context window.

Standard Intelligence 11-million-hour computer action dataset Devansh Pandey FDM-1 Galen Mead Sequoia Capital

Source coverage

Standard Intelligence: Training General Intelligence in Pixel Space Sequoia Standard Intelligence: Training General Intelligence in Pixel Space Sequoia

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.