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

Xie Saining describes himself as a "B-class" student who prioritized curiosity and personal interest over conventional academic metrics, leading him to unique opportunities and collaborations, including rejecting Microsoft Research Asia for an internship in Singapore.

He advocates for *World Models* as the fundamental building blocks for *general intelligence*, enabling prediction and planning in the *physical world*. He contrasts this with LLMs, which he views as primarily *communication tools* operating in digital space, not true thinking tools.

His journey was profoundly shaped by key figures like *Kaiming He* (who taught him non-linear research, "finding the gradient," and the importance of infrastructure), *Yang LeCun* (co-founder of AMI Labs, shared vision for World Models), and *Li Feifei* (defining problems).

Co-founded with *Yang LeCun*, AMI Labs aims to develop *predictive World Models* in a *research-friendly* and *open* environment, deliberately positioned outside the intense, LLM-centric competition of Silicon Valley. They are building a "grassroots alliance" to collect diverse real-world data.

Xie Saining views research as a *non-linear, exploratory process* driven by *curiosity* and the pursuit of *understanding*, rather than a finite game focused on immediate impact or accolades. He emphasizes the importance of *"Research Taste"* and questioning assumptions.