和@sainingxie 一起挑战7小时播客!他刚和Yann LeCun踏上“世界模型”的创业旅程(AMI Labs)。这是他第一次Podcast、第一次访谈。 2026年2月雪后的一天,我们在纽约布鲁克林,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。 这篇访谈的中文标题叫做《逃出硅谷》,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(侯晓迪、何恺明、杨立昆、李飞飞…)正是这些,让这篇“逃出硅谷”的对话充斥着人性的温度。 By the way, 下面是访谈的YouTube版本,我们提供了中英字幕。 And yes, 我们是在用播客给这个世界建模😎 A 7-hour podcast with Saining Xie. He has just begun a new journey on world models with Yann LeCun at AMI Labs. This was his first podcast appearance and his first long-form interview. A day after the snowfall in February 2026, in Brooklyn, New York, we started recording at 2 p.m. What followed became an unexpected marathon conversation that lasted until the early hours of the morning. The Chinese title of the interview is “Escaping Silicon Valley.” Yet throughout the conversation, he patiently listed the people who shaped his academic life, repeatedly sketching their personalities in vivid detail: Hou Xiaodi, Kaiming He, Yann LeCun, Fei-Fei Li, and others. These portraits are what give this “escape from Silicon Valley” conversation its human warmth. By the way, the YouTube version of the interview is below, with Chinese and English subtitles. And yes, we are using podcasts to model the world 😎 A 7-hour marathon interview with Saining Xie: World Models, AMI Labs, Ya... youtu.be/rIwgZWzUKm8?si=edxa… 来自 @YouTube

Mar 16, 2026 · 8:58 AM UTC

54
179
1,222
853,861
Sort replies: Relevant Recent Liked
7小时真的很惊人,要去两次厕所吧
1
12
11,905
Turned the full 7-hour podcast into notes for easier reading: lilys.ai/digest/8604396/9710…
1
10
5,029
赞👍 周末就来听 喜欢小珺的访谈
1
3,568
it was daylight when it started, it actually got dark at the end! that's how long it was!
1
4,609
7 hours?! Holy!
788
七个小时的播客!我需要在山野之间跑步50公里。
1
30
6,463
He rejected Ilya Sutskever. Twice. Not because he didn't respect the work — but because he believed in a different path.
10
3,684
Consider posting your videos/transcripts on Substack too? Super large English-speaking audience interested in China/tech there who are already bought into Substack tech podcasts (like Dwarkesh, Cheeky Pint...)
5
3,329
非常有启发的一期播客,和张祥雨那期一样精彩。几个感受:1、维特根斯坦的语言游戏,语言是一种精巧的人类建模,但也充满了限制,因为它首先是服务于交流的。2、通过LLM走向世界模型是有缺陷的,对真实环境的建模需要先投影到语言上,语言并不能完美表征环境,像素也不能,确实像拄着拐杖走路
1
2
987
这么长,播客的小别胜新婚。
2
2,206
看到停不下来的博客,好棒! 希望谢赛宁“导演”在AMI发更好的作品!
1
608
7 小时的对谈也太夸张了🙈 对听众的上下文要求好高啊……
1
1,592
I’ve just listened to the first three hours of the interview — it truly feels like an exciting marathon through Dr. Saining Xie’s research journey. What inspired me the most was how Dr. Kaiming He demonstrated the art of doing research to Saining, as well as the humility they both showed in preparing their papers. The funny part is that Kaiming introduced the Diamond Sutra (《金刚经》) to Saining when they were working at FAIR. I actually started reading the Diamond Sutra back in 2004 and even brought a copy with me when I moved to the United States. I still revisit it from time to time. There is a line in the sutra — “若见诸相非相,则见如来” — often translated as “If one sees that all forms are not truly forms, then one sees the Tathāgata.” When I used to write papers, I often felt that formatting and language were merely superficial appearances, while the underlying ideas were the real essence. After hearing how meticulously Kaiming refined the structure and wording of his papers, I realized that perhaps I had also been attached to the notion of being “beyond appearances.” Another great takeaway for me is that a top ML researcher also needs to be deeply grounded in AI infrastructure. Kaiming reportedly helped build large-scale training infrastructure spanning thousands of TPUs during his time at FAIR a few years ago. This really highlights his end-to-end understanding of AI — from systems and infrastructure, to model architecture, and ultimately to algorithmic innovation.
1
370
Saining Xie + Yann LeCun on world models is one of the most interesting directions right now — JEPA-style architectures that build internal representations of physical dynamics rather than predicting pixels. AMI Labs is a thesis bet that understanding precedes generation.
1
1,149
404 mins 太猛了吧
1
2,479
@grok help me understand his escape and current roadmap in three key takeaways 1. What to take 2. What not to take 3. Why he takes it
1
1
288
太强了,每次都在事件的第一线 👍
2,136
Lol. A whole lot of ambiguous BS. Met him irl once - no technical or mathematical skills to speak off.. but can yap a lot. Hung on the coatails of Lecunn from the get-go. Also, prefers grad students of a specific ethnicity. Terrible person
1
88
@zhang_benita is Lex Fridman of China
114
“你不要想着说你不做这件事,别人就会把这件事做了,你要想如果你不做这件事,这件事在这世界上永远不会发生” -- @sainingxie
1
3
899
“Always 去做更高维度的事情”
84
这期信息量很大,我的观点:真正稀缺的不是模型参数,而是‘把研究转成可交付产品’的组织能力。可执行建议:听完后各团队先写1页‘30天可上线功能清单’,别停在概念讨论。你最看好先落地到哪类场景?
189
居然穿的是 vivienne westwood 的衬衫,有点特别
328
7小时长访谈很稀缺,价值在“把思考过程完整暴露出来”。我建议再补一层结构化索引:按时间轴拆成「问题定义-技术路径-组织策略-融资判断」四段,读者会更容易二次传播。你们后续会考虑放出带时间戳的重点导航版本吗?
210
小珺越来越厉害了 也越来越漂亮了👍
560
不是很理解为什么要搞7个小时,我觉得就是纯粹自嗨
286
七小时可以听一整天了!
254
巨作。
127
LeCun转了,太长了怎么办?谁有TLDR?
708
可以采访到杨格吗? @TheGregYang xAI到底发生了什么?
2,667
小珺,得到了很多第一次
983
这是离职创业了么,我怎么记得去年的博客还是在腾讯呢?
1
2,643
超级马拉松
1,292
Thanks, Xiaojun. This was such a thoughtful and insightful seven-hour interview. I really learned a lot from it. I’ve published a four-part deep dive series on Substack based on the conversation. Here is the link: robonaissance.com/t/language… While working on it, I reshaped the narrative a bit, added some background context, and clarified a few of the more technical points. I hope it helps bring the interview to a wider global audience. I’d really love to hear your thoughts whenever you have time. And if anything feels off or raises concerns, please feel free to let me know. I’m also working in world models research, so it was especially meaningful to spend time with this conversation. I’d be excited to see more of your interviews in this space. Thank you again for putting together such a rich and generous discussion.
336