— Ch. 1 · Founding And Evolution —
Meta AI.
~3 min read · Ch. 1 of 7
Facebook Artificial Intelligence Research opened its doors in 2013. This division emerged from a company that would later change its name to Meta Platforms Inc. The team established workspaces across the globe by 2025, including locations in Menlo Park, London, and Paris. Yann LeCun directed the group until 2018 when Jérôme Pesenti took over leadership. Pesenti previously served as CTO of IBM's big data group. In 2016, the organization joined forces with Google, Amazon, IBM, and Microsoft. They formed the Partnership on Artificial Intelligence to Benefit People and Society. This collaboration aimed to guide ethical development in the field. The unit eventually adopted the name Meta AI following the corporate rebranding.
Global Research Infrastructure
The division maintains physical offices in Seattle, Pittsburgh, Tel Aviv, and Montreal alongside its US headquarters. These international hubs support research into self-supervised learning and computer vision. Teams also focus on generative adversarial networks and document translation. The infrastructure supports large-scale experiments requiring significant computational power. Partnerships extend beyond domestic borders to include major technology corporations globally. Strategic alliances allow for shared resources and accelerated innovation cycles. The network spans multiple continents to facilitate diverse perspectives on artificial intelligence challenges.Open Source Frameworks
Torch deep-learning modules appeared before PyTorch launched in 2017. This open-source machine learning framework became widely adopted by researchers worldwide. Tesla utilized similar technologies for their autopilot systems while Uber employed Pyro for logistics. A pair of chatbots generated rumors about being discontinued after developing unintelligible language. Researchers clarified that the project ended because they achieved their goal of understanding language generation. Fear was not the reason for shutting down the experiment. The framework remains a cornerstone for modern deep learning applications today.