Yann LeCun, the Turing Award-winning “godfather of deep learning” and Meta’s Chief AI Scientist since 2013, is reportedly set to leave the company in the coming months to found his own startup. This move, first detailed in reports from The Wall Street Journal and Financial Times, marks a seismic shift for one of Big Tech’s AI powerhouses, which has poured over $15 billion into AI this year alone. While LeCun hasn’t publicly confirmed, sources close to him indicate the decision stems from deep philosophical and strategic rifts.
Clues Indicating Why He Left
LeCun’s exit wasn’t abrupt, however, it’s been brewing for some time amid Meta’s evolving AI strategy.
Key Departure Factors:
- Philosophical Clash on AI’s Future: LeCun has long criticized large language models (LLMs) like those powering Meta’s Llama series as a “dead end” for achieving human-level AI (AGI). In speeches and interviews (e.g., his February 2025 AI Action Summit talk), he’s argued LLMs excel at text prediction but lack causal reasoning, physical world understanding, or true planning—capabilities even a “housecat” possesses effortlessly. Meta, under CEO Mark Zuckerberg, has doubled down on scaling LLMs for quick product wins (e.g., AI glasses, chatbots), diverging from LeCun’s emphasis on foundational research via his Joint Embedding Predictive Architecture (JEPA).
- Organizational Shifts: Meta’s AI division underwent its fourth restructuring in 2025, including a $15 billion acquisition of Scale AI and elevating 28-year-old Alexandr Wang (Scale’s co-founder) as head of applied AI—effectively LeCun’s boss. Another young hire, Shengjia Zhao, became a co-chief scientist focused on scaling breakthroughs. LeCun, 65, reportedly chafed at this “upstart” dynamic and resource reallocation away from his Fundamental AI Research (FAIR) lab. Recent layoffs (hundreds in AI) and a pivot from moonshot research to commercialization further sidelined his vision.
- Subtle Signals: LeCun’s X posts and talks grew increasingly vocal about LLM limitations (e.g., a 2024 interview calling scaling “complete BS” for AGI). Insiders note his frustration peaked post-Meta’s Q3 earnings, where Zuckerberg touted “superintelligence in sight” via LLMs—echoing rivals like OpenAI.
What He’s Planning Next
LeCun aims to launch a startup centered on “world models”—AI systems that simulate physical reality, predict action outcomes, and reason hierarchically using “energy functions” (optimization metrics for compatibility and cost). This builds on his Meta work (e.g., JEPA prototypes) but untethered from corporate timelines.
LeCun is reportedly already fundraising, receiving interest from VCs like Marc Andreessen, and will retain his NYU professorship for academic continuity. One can expect a focus on embodied AI for wearables/robots, emphasising efficiency over data-hungry scaling. It’s speculative and capital-intensive, but LeCun’s track record (pioneering convolutional neural nets) positions his next venture as a high-stakes bet on AGI’s “next paradigm.”
Possible Impact on Meta
This is a gut punch to Meta’s AI prestige. LeCun’s departure, alongside $40B+ annual CapEx, could signal talent flight risks, especially after poaching stars like Wang only to alienate veterans. FAIR could lose momentum on long-term innovation, forcing heavier reliance on LLM scaling (risking commoditization vs. OpenAI/Google).
Meta Stock dipped around 2% post-reports, reflecting investor jitters over strategy coherence. on the upside, it frees resources for products, but creates broader industry ripple. The move validates skepticism on LLMs, potentially accelerating “world model” investments elsewhere. X buzz calls it a “major moment,” with some hailing this as a “rebellion against AI scaling obsession.”
Time will tell, and knowing LeCun’s ability it will be an interesting space to watch.
