Is Scale 'Solving' Computer Vision? Researchers Clash Over 3D Understanding, Robotics Timeline
Sadlylifes41u · x · 2026-09-20
A debate over whether scale has effectively "solved" computer vision. Skeptics argue detection, segmentation, and tracking are only part of vision: 3D/4D scene understanding, physical state estimation, interaction understanding, prediction under intervention, and robust generalization remain open problems — limitations robots make especially visible.
The pro-scale side claims Astra one-shots all detection and segmentation tasks on any image, citing the Bitter Lesson: the same collapse happened to NLP in 2021, is now hitting vision, and they predict robotics within 5 years.
The core scientific question raised: which capabilities actually emerge with scale, where models still fail, and what role structure still plays.
Related event: Debate Erupts Over Whether Scaling Has 'Solved' Computer Vision(4 posts)→
More from AGI Musings
- If grant decisions are automatic, why not allow unlimited submissions? — PlisSergey · 2026-09-20
- EA was a loose forum-born movement that splintered; someone should write its ethnography — tszzl · 2026-09-20
- AI-speed grant review sparks debate: NIH's 6-submissions cap and the fairness puzzle — PTenigma · 2026-09-20
- Google researcher on AGI economists: useful near-term insights, post-AGI models are guesswork — docmilanfar · 2026-09-20
- When Opus 3 doubted base reality: LLMs and the surprisingly reasonable simulation argument — repligate · 2026-09-20
- Dev Builds Mnemos Cognitive Architecture and a "Retirement Home" for Deprecated AI Models — RileyRalmuto · 2026-09-20