LeCun: Entire AI Industry Is 'LLM-Pilled' in a Groupthink Trap
Yann LeCun, formerly of Meta and now at AMI Labs, has given a candid public assessment of the AI industry's structural dynamics: "The AI industry is completely LLM-pilled. Everybody is working on the same thing." He attributes this not to consensus belief but to competitive lock-in — companies are poaching each other's engineers at such rates that any deviation from the scaling trajectory risks falling behind while the pivot plays out. LeCun cites this dynamic as a reason he left Meta after it "became LLM-pilled." His core technical critique: agentic systems cannot work without world-model-based consequence prediction, and language fluency is "the easy part."
Why It Matters
LeCun's critique is gaining circulation precisely as the first architectural cracks appear — MILKYWAY showing scaffolding beats model upgrades, small models beating giants on structured tasks, and agent harness design emerging as the active frontier. His framing provides the structural argument for why alternative architectures are not being pursued, even when the evidence is mounting.