Kaiming He

Kaiming He appears 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 as one of Xie Saining’s strongest research influences at FAIR. The source connects him to ResNeXt, Self-Supervised Learning, MoCo, strong baselines, scalable models, and a disciplined research style.

Method

The interview emphasizes He’s focus, taste, engineering ability, reading selectivity, and habit of predicting experiment outcomes before running them. Xie treats this as a practical form of Research Taste: a good idea is not an isolated flash, but the result of input, exploration, baseline work, surprise, and pivot.

Connections

• Xie Saining — collaborator and narrator of the lessons.
• FAIR and Meta — research-lab context in the source.
• ResNeXt, Self-Supervised Learning, and Representation Learning — technical areas tied to his influence.
• Research Taste, Problem Definition In Research, and Frontier Model Scaling — methodological and scaling themes.