Xie Saining
Xie Saining is the AI researcher interviewed in 133. 对谢赛宁的7小时马拉松访谈：世界模型、逃出硅谷、AMI Labs、两次拒绝Ilya、杨立昆、李飞飞和42. The source traces his path through Shanghai Jiao Tong ACM training, a Singapore vision internship, UCSD PhD work, FAIR, NYU, and then AMI Labs with Yann LeCun.
Research Position
Xie frames his work as a long Representation Learning line rather than a set of isolated papers. In the interview, ResNeXt, Self-Supervised Learning, MoCo, Diffusion Transformers, REPA/RAE, Multimodal Intelligence, and World Models are all parts of the same search for useful abstractions over data, action, and physical reality.
Operating View
The source presents Xie as valuing Research Taste, strong baselines, infrastructure, credit assignment, and Problem Definition In Research. His two refusals of Ilya Sutskever and OpenAI are used to show that he prioritized route fit: first computer-vision research culture at FAIR, then a world-model route at NYU and AMI Labs.
Connections
- AMI Labs — company he is building with Yann LeCun.
- FAIR, Kaiming He, and Meta — research lab and collaborators that shaped his representation-learning work.
- NYU, Yann LeCun, and UCSD — university contexts in the source.
- Fei-Fei Li and ImageNet — influence on problem definition and visual intelligence.
- Ilya Sutskever and OpenAI — alternative route he twice declined.
- World Models, Joint Embedding Predictive Architecture, and Decentralized World Model Strategy — technical and startup direction he advocates.