Four years ago, French-American computer scientist Yann LeCun, then chief AI scientist at Meta, had presented an analogy of how a teenager who has never driven a car before can learn to drive in about 20 hours while the most advanced autonomous driving systems still require millions or billions of training data pieces and reinforcement learning trials.

Four years later, in the light of recent cybersecurity incidents by LLMs and the ongoing debate on “pacing the frontier”, LeCun has reiterated that autoregressive large language models (LLMs), which generate sequences by predicting the next token based on preceding context, will not achieve “human level intelligence.”

Taking to X social media platform on Sunday, LeCun said the current AI systems are are based non-auto-regressive search but they do it in token space, which he argued is limited and inefficient compared with the way humans and animals reason.

I said “auto-regressive LLMs, in and of themselves, will not lead human-level AI”
That statement is still totally true.

First, the reasoning abilities of current AI systems are based non-auto-regressive search (which is what I have always advocated for). But AFAICT, they do it…

— Yann LeCun (@ylecun) September 20, 2026

“if LLMs were a path to human-level AI, we would have domestic robots and Level-4 or Level-5 self-driving cars for consumers by now. And we don’t. We certainly don’t have cars that can learn to drive in 20 hours or practice like any teenager. We’re still missing something pretty huge to claim human-level intelligence (let alone superhuman),” wrote LeCun.

LeCun argues that today’s increasingly capable AI systems should not be mistaken for human-level general intelligence because benchmark performance and accumulated knowledge do not demonstrate the ability to rapidly learn, reason and adapt to genuinely unfamiliar situations, a capability he considers fundamental to intelligence.

“As Jean Piaget famously said- intelligence is not what you know, it is what you do when you don’t know. It is your ability to solve new problem without any prior training, to act in previously-unknown scenarios, and to adapt very quickly to new situations with minimal training. We’re still far from that,” added LeCun.

LeCun says that human-like reasoning would instead require AI systems to conduct searches in a continuous representation space, adding that the industry appears to be moving in that direction. He also questioned the effectiveness of current AI self-improvement methods, saying they work primarily in domains such as mathematics, coding and accurately simulated environments, where the quality of an AI system’s output can be evaluated without human intervention.

By contrast, he said humans and animals can learn new skills far more efficiently than current reinforcement-learning systems. That in crux is the idea behind Yann Lecun human-level AI.

LeCun also advocated for use of Joint Embedding Predictive Architectures (JEPA) trained through self-supervised learning, as a more effective approach to train AI systems.

Yann LeCun vs Geoffrey Hinton: Two Very Different Views on AI’s Future

Yann LeCun’s comments on X stemmed from a reply tweet he posted to Geoffrey Hinton, Noble prize laureate scientist widely considered as “godfather of AI”, three years ago, criticizing his statement that AI posed existential threat to humanity.

LeCun had accused Hinton of “helping those who want to put AI research and development under lock and key and protect their business by banning open research, open-source code, and open-access models.”

You and Yoshua are inadvertently helping those who want to put AI research and development under lock and key and protect their business by banning open research, open-source code, and open-access models.

This will inevitably lead to bad outcomes in the medium term.

— Yann LeCun (@ylecun) October 31, 2023

While Hinton is a known advocate of the AI existential threat theory, LeCun has mocked such threats claiming that current AI systems run on inefficient architecture to achieve anything even remotely close to human intelligence.

LeCun had recently suggested that “everyone should make fun of” Anthropic CEO Dario Amodei and other AI/tech CEOs who claim AI pose existential threat to humanity.

Also Read: Yann LeCun’s AMI Labs Raises $1.03 Billion to Develop “World Model” AI