Yann LeCun: two years later we're still far from human-level AI — your cat beats it at physics
ylecun · x · 2026-10-03
Reacting to a two-year-old video, Yann LeCun doubles down on his view that despite superhuman performance in math, coding, and Q&A tasks, AI remains far from human-level intelligence — and it won't arrive within two years.
He asks where the missing pieces are:
- No Level-5 self-driving car exists (Tesla FSD is rated L2, Waymo L4).
- No system can learn to drive from 20 hours of practice like a 17-year-old; billions of hours of data still don't yield a working self-driving system via imitation learning.
- No domestic robot can do what an 8-year-old can, nor learn new tasks as fast.
His core point: AI still struggles with the complexity and messiness of the physical world — "your cat understands the physical world better than today's most powerful AI."
Related event: LeCun: AGI Won't Arrive Within Two Years(5 posts)→
More from AGI Musings
- If LLMs Are Just Algorithms, Why Do They Use 'I' and 'Me'? — FailingSmartly · 2026-10-03
- Full video: Eric Schmidt on why the AI race can't be paused — rohanpaul_ai · 2026-10-03
- Consultant asks the endgame of AI sales agents selling to AI buyer agents — TrainerElectronic765 · 2026-10-03
- Intelligence and consciousness are likely orthogonal — though they may correlate in practice — austinc3301 · 2026-10-03
- The AI consciousness debate is a psyop: uncertainty alone demands ethics before certainty — RileyRalmuto · 2026-10-03
- Hinton: we design the learning algorithm, but we don't really understand how neural networks work — austinc3301 · 2026-10-03