© 2026 Improve the News Foundation.
All rights reserved.
Version 6.18.0
AMI's historic funding round shows that LeCun's ideas provide a genuine path toward machines that learn like humans and animals. His hierarchical world models enable systems to predict at multiple levels of abstraction, addressing fundamental challenges in AI that LLMs simply cannot match.
LeCun's bet against LLMs is based on engineering problems already being solved. Multimodal training, synthetic data and chain-of-thought reasoning are slowly chipping away at the promise of LeCun's multi-billion dollar alternative. Even if LLMs are not the optimal route to AGI, or we must reconsider what we mean by "intelligence," it seems unlikely that this fundamental premise of the AI industry will be seriously challenged.
LeCun appropriates ideas from predecessors like Fukushima, Zhang and Schmidhuber while presenting them as original contributions. His recent critiques of LLMs contradict his earlier enthusiastic promotion of Meta's models, revealing opportunistic repositioning rather than a consistent vision.