MIT's Vincent Sitzmann: traditional computer vision is about to go away
sreejan_kumar · x · 2026-10-05
MIT professor Vincent Sitzmann argues computer vision as we know it is about to disappear: intermediate tasks like segmentation and 3D reconstruction are historical artifacts, and world models already bypass them to solve far more general problems. He predicts vision's future lies in end-to-end perception-action loops—dissolving boundaries between vision, robot learning, and control—and that 3D representations will soon vanish from embodied intelligence, just as hand-crafted features did. sreejankumar's thread links the essay to closing the perception-action loop opening richer questions in control and adaptation.
More from AGI Musings
- Hinton still calls for AI safety with parent-baby analogy; compassion beats empathy — petitegeek · 2026-10-05
- Zed creator: 'What matters for humans matters for LLMs' — armchair takes aren't serious — zeeg · 2026-10-05
- The tech tree is less legible than you think: Moderna had a vaccine prototype before experts predicted — shakoistsLog · 2026-10-05
- repligate on AI optimism: models show compassion beyond what trainers intended — repligate · 2026-10-05
- AI should challenge you, not strictly obey you — HankYeomans · 2026-10-05
- If Meta's AI Agents each keep their own SQLite memories, how would CCPA data requests even work? — dbreunig · 2026-10-05