Photon-1 learns computer use from 18 years of unlabeled screen recordings
ycombinator · x · 2026-07-24
Y Combinator is spotlighting a stealth project called imagination models: a new foundation-model architecture designed to learn from internet-scale video.
- The first model, Photon-1, reportedly learned to use a computer by watching 18 years of screen-recording video with no action labels.
- The post positions this as a new way to train models from raw observational data rather than task annotations.
Related event: Induction Releases Photon-1 Learning Computer Use via Unlabeled Video(3 posts)→
More from Models
- Anthropic rumor says Opus 5 was scrapped after human-control concerns — haider1 · 2026-07-24
- Codex gets a usage-limit reset calendar for June–July 2026 — haider1 · 2026-07-24
- GLM-5.2 nearly doubles Kimi K3 on a long-horizon browser benchmark — zainhas · 2026-07-24
- KAT-Coder-V2.5-Dev goes open-weight with 35B total and 3B active parameters — AdinaYakup · 2026-07-24
- Opus 4.8 passes a biology question that 5.6 Sol xhigh misses — Sauers_ · 2026-07-24
- Identical weights can hide wildly different neuron utility — TheGradient · 2026-07-24