A 7-hour marathon interview with Saining Xie: World Models, AMI Labs, Yann LeCun, Fei-Fei Li, and 42

Video Description

In the spring of 2026, as China celebrated the Chinese New Year with humanoid robots dancing on the national stage, New York was reeling from a brutal blizzard—the harshest winter the city had seen in years. In a cluttered apartment in Brooklyn, my conversation with Saining Xie began. By March 2026, Xie, alongside Turing Award laureate Yann LeCun, had embarked on a high-stakes venture into "World Models." Their brainchild, AMI Labs (Advanced Machine Intelligence Labs), consists of just 25 people and zero products. Yet, it has already secured a $1.03 billion Seed round at a pre-money valuation of $3.5 billion. This is the titan among 2026’s "AI Neo Labs," setting a record for the largest Seed round in European history. Statistically, its initial funding ranks second globally among Neo Labs, trailing only Thinking Machines Lab, founded by former OpenAI CTO Mira Murati ($2 billion Seed). "Silicon Valley is very LLM-pilled," says Xie, AMI’s co-founder and Chief Science Officer. "The Valley is deep in the trenches of Large Language Models—it’s completely hypnotized." Born in 1990, the Chinese scientist is an alumnus of Shanghai Jiao Tong University and UC San Diego, and currently holds a faculty position at NYU. Before venturing into the startup world, he was a Research Scientist at Google DeepMind and spent four years at Meta’s FAIR. With nearly 100,000 citations to his name, he is best known for co-authoring Diffusion Transformers (DiT). This is Xie’s first-ever in-depth interview. On a snowy afternoon in February 2026, we began what turned into an unexpected marathon, starting at 2 PM and only dispersing in the early hours of the morning. Under the neon-soaked, pitch-black streets of New York, Xie’s rhetoric echoed the city’s own grit—a complex blend of sharpness and nuance. When it comes to the prevailing AI dogma, he doesn't pull his punches: - "I don’t see LLMs as a success story for the 'Bitter Lesson'—in fact, they are a rebellion against it." - "There is significant 'water'—inflation—in the Scaling Laws of language models." - "Language is 'poison,' or perhaps, an 'opium.'" - "I am deeply concerned about the 'pollution' of vision by language. Actually, it’s already happening." This dialogue is a complete odyssey through the technological upheavals of the last 15 years and the people who forged them. Often, when the topic drifted, Xie insisted on circling back to finish his "portraits"—perhaps because his original dream was to be a film director. While that dream was shattered early on, it has found a new form of expression here. It is this very quality that makes this "anti-OpenAI," "anti-Silicon Valley," and "anti-hegemony" dialogue feel so profoundly human. In this world, he says, "Only the sincere exchange between human beings matters. Perhaps nothing else does." Disclaimer: This content does not constitute investment advice. --- 2026年春节，在中国机器人登上春晚的喜乐时分，纽约刚下过一场暴雪——这是近几年以来纽约最凛冽的一个冬天。在布鲁克林一栋略显凌乱的楼房，我与谢赛宁开始了这场对话。 2026年3月，他与图灵奖得主杨立昆（Yann LeCun）等，踏上了一条关于“世界模型”的创业旅程。他们创立的AMI Labs（Advanced Machine Intelligence Labs），目前仅25人、在没有任何产品的情况下，完成10.3亿美元Seed轮融资，投前估值35亿美元。 这是2026年新创的AI Neo Labs中备受瞩目的一家，创造了欧洲历史最大Seed轮。从统计看，它的第一笔融资在Neo Labs中位居世界第二，仅次于前OpenAI CTO Mira Murati创立的Thinking Machines Lab（20亿美元Seed轮融资）。 “Silicon Valley is very LLM-pilled。”AMI联合创始人兼首席科学官谢赛宁说道，“硅谷已经深陷于LLM（大语言模型），完全被它催眠了。” 这位出生于1990年的华人科学家，毕业于上海交通大学与加利福尼亚大学圣地亚哥分校，现于纽约大学任教。创业之前，他曾在Google DeepMind担任研究科学家。更早之前，他在Meta的FAIR实验室担任研究科学家4年。他的论文总计引用数近10万次，曾共同提出Diffusion Transformers（DiT）。 这是谢赛宁第一次接受访谈。2026年2月雪后的一天，我们从下午2点，开启了一场始料未及的马拉松式访谈，直到凌晨时分散去。 在黑暗浸透的霓虹纽约街头，空气中弥漫着没化完冰雪与难闻烟雾的混合味道。就像谢赛宁的表达，总是带着多重的混合感。 在人工智能的思想上，他不乏尖锐言论： - “我完全不觉得LLM是Bitter Lesson（专指Richard Sutton提出的《苦涩的教训》）的成功展示，某种程度上，LLM是反Bitter Lesson的”； - “语言模型的Scaling Law里面是有水分的”； - “语言是‘毒药’，或者说，是一剂‘鸦片’”； - “我非常担心语言对视觉的污染，事实上，它已经发生了”。 但他又不厌其烦地，枚举了影响他学术生涯的每一个人，并反反复复口头描摹这些人的人物特征。 这里面有：写《交大生存手册》的侯晓迪，初来乍到美国还不会用Linux的何恺明，活得像65岁青春期少年的杨立昆，捱过许多艰难时刻、以勇气与坚持让AI看见世界的李飞飞，等等。 这场对话涵盖了过去15年间，他对于AI技术变革兼塑造这些变革背后的人的完整游历。很多次我们的话题早就转移了，他坚持要绕回来，把人物素描完成——也许是因为，他人生最初的梦想是成为一名导演，尽管这个梦想很快就破灭了，却以一种新的形式延续下去。 但正是这些，让这篇“反OpenAI”、“反硅谷”、“反霸权”的对话，又充斥着人性的温度。 他说，在这个世界上，“只有人与人之间真诚的交流是重要的，也许其他都不重要”。 免责声明：本内容不作为投资建议。 --- OUTLINE: 00:01:19 The Normal One —— 普通人 00:35:40 The World Won't Let Me Do Vision —— 世界不让我做视觉研究 00:52:06 Academic Nomadism —— 学术流浪 00:57:43 A Friendship with Kaiming He —— 与何恺明的友谊 01:05:35 Turning Down Ilya Twice —— 两次拒绝Ilya 01:08:26 Memories of Yann LeCun and Fei-Fei Li —— 杨立昆与李飞飞往事 01:12:18 Hidden Clues: A World of Representations —— 隐藏的线索：表征的世界 02:43:55 Research Taste and the Diamond Sutra —— 研究品味与《金刚经》 04:11:07 What is a World Model? —— 什么是世界模型？ 04:29:47 From Downloading the Internet to Downloading Humanity —— 从下载互联网，到下载人类 04:58:17 The Genesis of AMI Labs with Yann LeCun —— 与杨立昆共同创立AMI始末 05:45:53 "Silicon Valley is Hypnotized" —— "硅谷已被催眠了" 06:07:17 The Arrogance of Mankind —— 人类的自大！ 06:18:45 "42" —— "42"