After twelve years with Meta, including five as founding director of the Facebook AI Research (FAIR) lab and seven as Chief AI Scientist, Yann LeCun is leaving to launch a startup focused on the Advanced Machine Intelligence (AMI) research program. This program, pursued with colleagues at FAIR and New York University, aims to develop AI systems capable of understanding the physical world, utilizing persistent memory, and performing complex reasoning and planning. The creation of FAIR itself is considered LeCun’s most significant non-technical achievement, and Meta will partner with the new company to maximize AMI’s broad impact across multiple sectors.
Yann LeCun’s Departure from Meta
Yann LeCun plans to leave Meta after 12 years, including five as founding director of FAIR and seven as Chief AI Scientist. He intends to create a startup focused on Advanced Machine Intelligence (AMI) research, building on work done at FAIR, NYU, and beyond. Meta will partner with the new company, recognising potential overlap with its commercial interests, but also opportunities beyond them. LeCun believes pursuing AMI independently will maximize its broader impact.
A key shift began in 2018 when LeCun moved to an individual contributor role, coinciding with the rise of transformer technology stemming from Google’s 2017 paper. By 2020, transformers began surpassing Convolutional Neural Networks (CNNs) – the foundation of LeCun’s legacy in deep learning. This led him to focus on world models, JEPA, and AMI, putting him at odds with Meta’s focus on large language models and billions of dollars in investment.
The source highlights a split emerging in AI: “capability-first models” (token-predictors) versus “coherence-first systems” (persistent world models). LeCun’s work is geared towards the latter, seeking systems that maintain meaning across time, memory, reasoning and action. The next frontier, according to commentary, isn’t bigger models, but ensuring semantic grounding while machines learn and act, emphasising auditability of coherence over simple capability.
Focus on Advanced Machine Intelligence (AMI)
Yann LeCun is launching a startup to pursue Advanced Machine Intelligence (AMI) research, building upon work done at FAIR, NYU, and beyond. The goal of AMI is to create AI systems that can understand the physical world, possess persistent memory, reason effectively, and plan complex action sequences. Meta will partner with the new company, recognizing the potential for AMI to have broad applications, including some that overlap with, but also extend beyond, Meta’s current commercial interests.
The pursuit of AMI represents a shift in AI focus, moving beyond simply scaling up current models. A key distinction is emerging between “capability-first” models – like large language models – and “coherence-first” systems focused on persistent world models. The next wave of AI will prioritize maintaining meaning across time, memory, reasoning, and action, rather than solely generating more language or achieving higher error rates.
A core focus of AMI will be creating AI that is auditable, ensuring coherence between intention, state, and behavior. This means developing systems capable of verifying meaning continuously throughout their lifecycle, not just evaluating output. This is critical for creating machines that can plan, reason, and remain aligned with the world they operate in—a departure from models simply focused on capability and scale.
