Meta's Yann LeCun Says AI Still Can't Match Human Learning

LeCun highlights the limitations of large language models and the gap between AI and real-world intelligence.

Published on Feb. 27, 2026

Got story updates? Submit your updates here. ›

Former Meta AI chief Yann LeCun said at the India AI Impact Summit 2026 that while large language models are "incredibly useful," artificial intelligence still struggles with tasks humans find simple, like driving. LeCun compared LLMs to a modern evolution of the printing press, libraries and the internet, noting that AI excels in information retrieval and symbolic reasoning but lacks a true understanding of the physical world. He stressed that humans and animals learn by observation and interaction, building "mental models" to predict outcomes and adapt to new situations, which AI has yet to achieve.

Why it matters

LeCun's comments highlight the ongoing challenges in developing AI systems that can match human-level learning and adaptability, particularly in real-world scenarios. As AI becomes more prevalent, understanding its limitations is crucial for setting realistic expectations and guiding future research and development.

The details

LeCun noted that while AI can pass the bar exam and excel in math Olympiads, "We certainly do not have self-driving cars that can teach themselves to drive in 20 hours of practice, like a 17-year-old." He explained that AI, in contrast to humans and animals, cannot yet navigate the messy, unpredictable real world, making robots and self-driving cars less capable than human learners. LeCun also highlighted the potential of AI as a tool to amplify human intelligence and improve access to knowledge, similar to the impact of the printing press centuries ago.

  • The India AI Impact Summit 2026 took place on February 20, 2026.

The players

Yann LeCun

A leading figure in artificial intelligence, machine learning and robotics, and a professor at New York University and the executive chairman of AMI Labs. He is also an ACM Turing Award laureate.

Meta Platforms, Inc.

The parent company of Facebook, which LeCun previously served as the chief AI scientist.

Got photos? Submit your photos here. ›

What they’re saying

“We certainly do not have self-driving cars that can teach themselves to drive in 20 hours of practice, like a 17-year-old. We're missing something big.”

— Yann LeCun (India AI Impact Summit 2026)

“It's just a more efficient way to access information.”

— Yann LeCun (India AI Impact Summit 2026)

What’s next

LeCun's comments suggest that further research and development is needed to bridge the gap between AI and human-level learning, particularly in real-world scenarios. This could involve exploring new approaches to machine learning and building more robust 'mental models' that can better adapt to unpredictable environments.

The takeaway

While large language models and other AI tools have made significant advancements, they still fall short of matching the learning capabilities of the human brain. Recognizing these limitations is crucial for setting realistic expectations and guiding the future of AI research and development to unlock its full potential as a tool to amplify human intelligence.