Kevin

Coder, Product Manager, Investor and Entrepreneur
Co-Founder&CPO @codatta_io
Partner @Zen Capital
Investor @AlibabaGroup
Product @AntGroup
Architect @Oracle

[Kevin @qiw]

#59: Finished Lee Smolin's Einstein's Unfinished Revolution. Quantum weirdness isn't the end — it's a doorway to deeper reality. The revolution Einstein started may still be physics' true future.

[Kevin @qiw]

微信读书第59本：李·斯莫林《量子力学的真相》，量子世界的诡异不是终点，而是通往更深实在的入口。爱因斯坦未竟的革命，或许才是物理学真正的未来。

[Kevin @qiw]

Training the future of autonomous farming. Check out this high-quality dataset of robot tea picking from a POV perspective. 🤖🍵 #AgTech #Robotics #AI #DataScience

[Kevin @qiw]

数据要产生价值需要闭环：采集→标注→训练→部署→再采集。飞轮转速取决于两个变量：机器人的数量，和任务的复杂度。宇树在量上有优势，特斯拉优势在于Optimus执行的是极度标准化、高重复性、且特斯拉完全掌控的任务。这意味着特斯拉可以极快地迭代，因为它同时控制了硬件、软件、部署场景和数据标注

[Kevin @qiw]

语言模型可以从互联网上抓取数万亿个文本token；而具身智能需要的是"机器人在真实物理环境中执行任务"的数据-这种数据只能由实际运行的机器人产生。
宇树现在有超过3万台四足机器人和5500台人形机器人在全球运行。这是一个大规模物理数据采集网络正在成型的信号。

[Kevin @qiw]

What a four-year-old does without thinking may be the field's hardest unsolved problem. If that inversion holds, we're not just misallocating resources. We're optimizing for the wrong definition of intelligence entirely

[Kevin @qiw]

Language as a training signal has a structural flaw that scale can't fix — the data isn't the world, it's humanity's filtered description of the world. Most LLM criticism stays behavioral: hallucinations, reasoning failures. @sainingxie 's critique cuts deeper.

[Kevin @qiw]

And underneath all of this is a harder question about benchmarks. The tasks we've built the industry around — olympiad math, bar exams, coding contests — reflect what impresses humans, not what's computationally fundamental.

[Kevin @qiw]

现在整个行业的benchmark体系是围绕人类认为困难的任务建立的，但人类认为困难的事情，对计算机来说不一定难；人类认为trivial的事情，比如在真实物理环境里生存，反而可能是更根本的挑战。这个认知翻转如果是对的，整个行业的资源配置方向就是错的。

[Kevin @qiw]

"去中心化联盟对抗LLM霸权"这个叙事觉得有点过于整洁。现实中数据拥有者的动机是复杂的，有人想要回报，有人怕竞争对手拿到自己的数据，有人根本不知道自己的数据有什么价值。这些摩擦在访谈里没涉及但却是实操时最现实的问题。

[张小珺 Xiaojun Zhang @zhang_benita]

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... https://t.co/3rTwdTGkJI 来自 @YouTube

[Kevin @qiw]

Transformer处理连续高维信号的架构局限也是真实的。对每个token付出相等注意力这件事，在语言里是合理的，但放到视频、工业数据上这个假设可能就错了。

"从真实世界收集数据训练世界模型"这个思路方向没问题，但访谈里对数据获取的复杂性处理得很轻。工业传感器数据、医疗数据等的获取不只是商业合作

[Kevin @qiw]

Thanks for the invitation and looking fwd to the discussion~

[FUTUREMODE 台灣未來祭 (prev Taipei Blockchain Week) @TaipeiWeek]

Building the future of decentralized data.
Kevin Wang (@qiw), Co-founder & CPO of @codatta_io, brings experience from Alibaba, Ant Group, and Oracle to shape the next generation of AI x Web3 infrastructure.

Meet him at #TBW2025.
📍 Songshan Cultural and Creative Park | Sep 4–6th, 2025
🎟️ https://t.co/4Fka50Kgwx

[Kevin @qiw]

Honor to make our contributions to open source AI Training data sets, more to come~ stay turned

[Codatta @codatta_io]

See a million plates from every corner of the world, named, tagged, and verified by thousands, and trace each one back to its source... That's MM-Food-100K. Built with 87,000+ contributors over six weeks, every entry is linked to a wallet and refined through a hybrid workflow of AI drafts + human experts.

The full corpus holds 1.2M quality-accepted samples; 100K (a curated 10%) is on @huggingface for non-commercial use, with the remaining 90% reserved for commercial licensing and royalties to contributors.

Why does it matter? Robotics can recognize your plate, smart kitchens know what you cook, supply chains track what moves, and health apps turn images into nutrition insights.

Powered by the @Binance community, you turned individual submissions into a living map of global food knowledge. So if you're in health & nutrition — think @MyFitnessPal, @noom, @Lifesum — let's talk! Your apps deserve food data with traceable provenance.

Food data with provenance: open for experimentation, and protected where it drives value.

Built with the crowd, and for the world.

📄 More on the dataset: https://t.co/ohhXfj5CVd
📂 Access it here: https://t.co/wOmHq3Qcfa

[Kevin @qiw]

Thanks for @GoKiteAI and @avax super support 🙏

[KITE AI @GoKiteAI]

As part of the @GoKiteAI ecosystem incubation program, @codatta_io received funding from the @AvalancheFDN ’s InfraBUIDL (AI) initiative, a $15 million grant program in which Kite AI serves as a joint selection committee member.

The InfraBUIDL (AI) program is designed to accelerate innovation at the intersection of AI and decentralized infrastructure. This funding not only provided critical resources but also connected Codatta to Kite AI’s broader ecosystem, accelerating both technical deployment and ecosystem collaboration.

With this support, Codatta achieved key milestones such as launching the first autonomous medical data annotation workflow, combining blockchain’s auditability with AI’s problem-solving capabilities and highlighting the strong demand for high-value AI assets.🌐

[Kevin @qiw]

Lower development costs, higher capital efficiency, and faster profitability. Seed-Strapping: Secure $100K-$1M in seed funding from investors focused on early revenue generation and strong returns. Ideal for startups leveraging AI for growth. #SeedStrapping

[Kevin @qiw]

A vast eval ecosystem for assessment tasks will emerge to customize Agents for every enterprise. #DataAnnotation #AI #FutureOfWork

[Kevin @qiw]

The data annotation market is undergoing a huge transformation. It used to rely on the crowdsourcing model with low - to mid - skilled workers. Now, it's shifting to "filtering" high - level Q&A, seeking the world's top talents to push the boundaries of model capabilities. #evals

[Kevin @qiw]

Previously, most evaluation tasks were "zero - shot" tests. But what we truly need to assess is "work valuable in results." In the future, Agents will take on various roles currently performed by employees, working with human staff, who will also help build these evaluation tasks