➤ This is a rare public discussion — one without convergence — on fundamental disagreements over the future path of embodied AI.
In 2026, embodied AI has no shortage of hype. What's missing is people putting the disagreements on the table. Where does the data come from? How do you build the models? Where does it actually land?
We discussed the hottest topics of the moment: Where are the capability boundaries of GPT-6 Astra? If OpenAI or Anthropic enter embodied AI, would that be a dimensionality-reduction strike on the whole industry?
This episode was recorded live at the main-stage roundtable of the "2026 Inclusion Bund Conference," hosted by me (Koji), with four guests:
Zheng Han: Co-founder and CEO of Sudu Tech
Yujun Shen: Chief Scientist at Ant Group's Lingbo Technology
Qian Wang: Founder and CEO of Independent Variable Robotics
Huazhe Xu: Founder of Poke Robotics, Assistant Professor at the Institute for Interdisciplinary Information Sciences, Tsinghua University
Through this conversation, you'll learn: 1) the core assumptions and risks behind different technical approaches; 2) how to identify which metrics are leading indicators; 3) the impact and inspiration Astra brings to embodied AI, and the moats available to startups; 4) the short-term vs. long-term logic of embodied AI commercialization; 5) which areas are overhyped or underrated — and the corresponding investment opportunities or risks.
🎬 Our video podcast will be released simultaneously on @Koji杨远骋's WeChat Channels, Douyin, Xiaohongshu, Bilibili, YouTube, and other platforms.
📒 The text version will be published on the @十字路口Crossing WeChat official account.
🟢 The Data Debate: Between Simulation and Reality
00:27 Simulation vs. real robots: it's not either-or
04:00 The imperfection of real-world sensors
06:14 Real-to-Sim: the harder gap
🟢 The Intelligence Debate: Astra's Impact and Lessons
08:18 GPT-6 Astra: a dimensionality-reduction strike on the embodied AI industry?
09:21 "The embodiment of intelligence lives in the data"
15:00 Mapo tofu doesn't prove the model got better
19:08 Embodiment will, in turn, change large models
23:01 High success rates and generalization: you need both
🟢 The Scenario Debate: The Metric That Crosses the Bubble
24:48 The only metric that crosses the bubble?
25:44 High success rates are the prerequisite for commercialization
26:55 Customers who keep paying
29:11 Post-training deployment and foundation model leaps
🟢 Looking Back in Five Years: What's Overhyped, What's Underrated?
31:16 Five years from now, what will look overhyped?
33:27 Deployment is a systems engineering challenge
35:01 Embodied AI won't replicate the language model playbook
38:33 Without physical replication, there's no true RSI
40:19 Building airplanes, the way we build apps today
41:25 The myth of a million hours of data
Welcome to subscribe to "Crossing":
🚦 We follow the new industry changes and startup opportunities brought by the new wave of AI technology.
🚦 "Crossing" comes from Steve Jobs' metaphor for Apple — standing at the intersection of technology and the humanities, where great products are often born. AI is transforming every industry. We seek out, interview, and bring together a new generation of AI founders and active builders of the AI era, exploring and embracing new changes and new possibilities together with them.
👦🏻 Host Koji: I founded Crossing and launched AI Hacker House, a community space for the new generation of AI founders. I'm also a Venture Partner at ZhenFund. I believe technology — especially AI — is our generation's greatest opportunity for value creation. Koji on Jike, Koji's website
View the transcript of this episode on Xiaoyuzhou