Saining Xie
Co-founder & CSO, AMI Labs
About
Saining Xie (谢赛宁) is co-founder and Chief Science Officer at AMI Labs, a company he co-founded with Yann LeCun to build world models that go beyond large language models. He is also an assistant professor at NYU's Courant Institute on leave. Xie received his Ph.D. from UC San Diego, spent four years as a research scientist at Facebook AI Research (FAIR), and later joined Google DeepMind's GenAI team. His research focuses on scalable visual representation learning, generative modeling, and multimodal understanding. AMI Labs raised a $1.03B seed round in March 2026.
Key Contributions
- Introduced ResNeXt, showing how grouped convolutions and cardinality could improve deep vision backbones
- Created ConvNeXt, demonstrating that carefully modernized convnets could compete with Vision Transformers
- Co-authored Masked Autoencoders with Kaiming He and colleagues, helping revive self-supervised visual representation learning
- Co-authored Diffusion Transformers, a design line later associated with high-end video generation systems such as Sora
- Co-founded AMI Labs with Yann LeCun to push world-model research beyond language-only systems
- His work shows that architecture details still matter in the transformer era, though popular accounts can over-credit single papers for later product systems
Videos & Interviews
Connections
Yann LeCun
CollaboratedExecutive Chairman & Co-founder, AMI Labs; Professor, NYU
Xie was LeCun's colleague in Meta's FAIR before becoming co-founder and chief science officer of AMI Labs, the Paris company LeCun started in December 2025 after leaving Meta over its bet on language models. They are building world models — systems trained on physical reality rather than text — around LeCun's JEPA architecture. It is the most heavily funded institutional wager yet that the current paradigm is a detour.
en.wikipedia.org · techcrunch.com
Fei-Fei Li
CollaboratedCo-founder & CEO, World Labs; Special Advisor on AI, Stanford
Both names sit on Cambrian-S (November 2025), the NYU paper that frames spatial supersensing as four stages beyond language — naming what is seen, remembering a stream, inferring the 3D world behind the pixels, predicting it — with LeCun beside them on the author list. In his seven-hour interview Xie says Li 'helped us with a great deal of valuable advice' on it, and names her among the people who 'paved the road' for Chinese vision researchers before him. It is the rare edge where the two rival world-model bets, AMI Labs and World Labs, share a byline.
arxiv.org · youtube.com
Immanuel Kant
KindredPhilosopher of the Enlightenment
In his seven-hour interview — on this site — Xie reaches for Kant's thing-in-itself, alongside Schopenhauer and the Diamond Sutra Kaiming He gave him, to explain research taste: what you see is not the essence, so break through a paper's surface and ask what actually lies beneath it. The reflex runs deeper than one citation — his world-model program is a wager that intelligence itself means modeling what hides behind appearances, a builder's answer to a question Kant ruled unanswerable.
youtube.com
