Yann LeCun

On 25 September 2026, Yann LeCun answered the economist Noah Smith on X: "Nope. It's still true." Smith had called LeCun's position already refuted, that position being that scaling LLMs will not reach human-level intelligence and that the idea of a "country of geniuses in a data center" was "complete BS."

LeCun's defence was a list of absences. Where is your domestic robot? Your level-5 autonomous car? Your robot car capable of learning to drive in a few hours of practice, like any 17-year-old? Your AI system able to understand the real world and pick up new skills quickly, like a house cat? The post carries no figure, no timeline, and no description of the alternative architecture. Its argument rests on learning efficiency.

He holds two claims at once, which the public debate tends to collapse into one. AI and computing already beat humans at a growing number of tasks. AI also remains far from general human intelligence, and when it arrives it will not be built on LLMs, which he expects to survive as a text interface. He had made the same point on 20 September, in a reply addressed partly to Geoffrey Hinton.

LeCun founded and chairs AMI Labs, teaches at NYU, and was Chief AI Scientist at Meta. He also backs ZML, the Paris startup behind the LLMD inference server.