Yann LeCun at ETH Zürich: Scaling LLMs to reach AGI is 'impossible'
rohanpaul_ai · x · 2026-10-03
In his latest talk at ETH Zürich, Yann LeCun argued that scaling LLMs to AGI is "impossible." His key comparison: an LLM trains on 30 trillion tokens (10^14 bytes of text, which would take a human 400,000 years to read), while a 4-year-old absorbs the same 10^14 bytes through vision alone in about 1 year and 10 months. He defines intelligence as the ability to quickly learn new tasks without prior training — a teenager learns to drive in 20 hours — and contends that scaling only increases stored knowledge, not this adaptive capability.
More from AGI Musings
- RIP traditional data scientists: the role is splitting into two tracks — mdancho84 · 2026-10-03
- David Patterson: solar's near-vertical cost curve will power the singularity — davidpattersonx · 2026-10-03
- ryolu_: When everyone copies what works, we stop making anything of our own — ryolu_ · 2026-10-03
- VC accuses OpenAI of skynet-style regulatory capture, says Anthropic's version is more subtle — StewartalsopIII · 2026-10-03
- Peking University mathematician: AI's millennium-prize wins are 'harvesting,' not discovery — 新智元 · 2026-10-03
- Anil Seth pushes back on Steve Yegge: denying machine consciousness isn't human exceptionalism — anilkseth · 2026-10-03