CR-T03
Not yetDisputed
"There's absolutely no way … that autoregressive LLMs, the type that we know today, will reach human intelligence." Scaling "is saturating".
- Claim made
- Who
- Yann LeCun (then Meta)
- Target date
- None stated
- Status checked
- 28 September 2026
Source
PYMNTS report [F] (secondary report of a talk)
Evidence to date
Evidence cuts both ways. For continued progress: agent success on Terminal-Bench 20%→77.3%; frontier models meet or exceed human baselines on PhD-level science questions (AI Index 2026 takeaways). For limits: models still struggle with multi-step planning and read analog clocks correctly 50.1% of the time (AI Index 2026 report page).
Next indicator
Whether leading labs move away from autoregressive LLMs; progress on long-horizon and physical-world tasks