Yan LeCun, a pioneering AI scientist and Turing Award winner, has left Meta due to disagreements over the company’s shift from fundamental AI research to rapidly deploying large language models under new leadership. LeCun, skeptical of the current LLM-focused approach, is now pursuing alternative AI methods through his new project VJEP, potentially challenging Meta’s AI strategy and future competitiveness.
The video discusses the recent departure of Yan LeCun, one of the world’s foremost AI scientists and a Turing Award winner, from Meta’s AI research lab. Yan LeCun is renowned for pioneering convolutional neural networks and deep learning, foundational technologies behind computer vision, facial recognition, and self-driving cars. His exit from Meta is significant because he has been a leading figure in AI research since 2013, and his departure signals potential trouble for Meta’s AI ambitions, especially as the company has struggled to keep pace with competitors in the AI race.
A key reason for LeCun’s departure is a shift in Meta’s AI strategy under Mark Zuckerberg, who has pivoted away from long-term fundamental AI research toward rapidly deploying large language models (LLMs) and AI products. LeCun reportedly had to report to Alexander Wang, a younger executive without comparable academic credentials or research stature, which LeCun saw as a downgrade and a loss of scientific autonomy. This change in leadership structure and focus conflicted with LeCun’s vision and values, contributing to his decision to leave.
LeCun is also a vocal skeptic of the current emphasis on large language models as the path to artificial general intelligence (AGI). He argues that LLMs, while useful, lack the ability to reason and plan like humans and represent an “off-ramp” rather than a true path to AGI. This skepticism is reinforced by Meta’s recent struggles with the Llama 4 model, which was widely regarded as a failure due to poor performance and ethical controversies. LeCun distanced himself from the Llama models, emphasizing that he was not involved in their development and criticizing the industry’s overreliance on generative models.
Instead of focusing on LLMs, LeCun advocates for a different approach to AI development, emphasizing joint embedding predictive architectures, energy-based models, and predictive model control over generative models and reinforcement learning. He is now working on a new project called VJEP, which aims to build AI systems that understand, predict, and plan in ways inspired by human and animal mental models. This approach focuses on creating world models that can reason and plan efficiently, potentially leading to more advanced and human-like AI.
The video concludes by reflecting on the implications of LeCun’s departure for Meta, which has recently laid off hundreds of AI researchers and struggled to produce competitive AI products. While Meta continues to invest heavily in AI, its direction appears uncertain, and losing a figure like LeCun could hinder its long-term prospects. Meanwhile, LeCun’s new startup is expected to attract significant investment and could become a major player in the AI field, potentially forcing Meta to reconsider its strategy or even acquire his new company in the future to stay competitive.