➤ This week's Crossing guest is Mengdi Xu, an assistant professor at Tsinghua University's Institute for Interdisciplinary Information Sciences (IIIS). Starting from vehicle engineering at Tsinghua, with stops at Johns Hopkins and CMU, she then did her postdoc at Stanford with Jiajun Wu and Fei-Fei Li's groups. In early 2024, she returned to China to join Tsinghua IIIS. Her core research focus is in-context learning (ICL) for robots: getting a robot to enter a new environment and learn a new task on the spot through just one or two interactions — and get faster at learning as it goes.
This episode also has a timely postscript: the week after we recorded, Generalist released GEN-1.5 — showing robots a 3-12 second motion demonstration as a "physical prompt," with no fine-tuning or parameter updates, and having them attempt the new task immediately (averaging ~59% success across 10 tasks). ICL instantly became one of the hottest topics in global embodied AI — exactly the direction Mengdi has been betting on.
Beyond ICL, we also discussed:
- The ceiling of pretraining
- True generalization
- Signals and traps of Scaling Laws
- Building stable data flywheels
- The misleading nature of folding laundry demos
- Industry bubbles vs. genuine progress
- The real life of a young PI and research choices
Beyond all the technical topics, this remains a conversation about life choices, environment, and long-term thinking — hopefully a reference for all young people standing between China and the US, academia and industry, following trends and holding the long line.
🎬 Our video podcast will be available simultaneously on Koji's WeChat Channels, Douyin, Xiaohongshu, Bilibili, YouTube, and other platforms.
📒 The transcript will be published on the CrossingCrossing WeChat official account.
🟢 From RC Cars to Robots
00:21 Rapid-fire Q&A: age, alma maters, MBTI, zodiac sign, one-line self-introduction
02:31 Tsinghua vehicle engineering, the starting point toward robots
05:59 The girl who scavenged scrap metal for RC cars
07:18 Hengshui High School, the other side of the competition class
12:02 CMU's Best Dissertation Award: beyond pretraining
19:19 Why is adaptation getting more attention?
🟢 From Problem-Solver to Problem-Poser
23:51 Creative tool use: how do you cut a cake without a knife?
26:30 Stanford: Fei-Fei Li and Jiajun Wu, two types of mentors
30:57 From Stanford back to Tsinghua
34:17 In 5 years, what will make you a top researcher?
40:06 Choose the best, or choose the right fit?
42:24 What kind of cold emails would she actually respond to?
🟢 Signals of Embodied Scaling Laws
44:44 Embodied startups: progress and bubbles coexisting
46:39 Data and world models, why does the wind keep shifting?
49:07 Low loss doesn't equal high success rate
50:47 Are robots still at the GPT-1 stage?
56:04 In-Context Learning: can a robot understand your home?
58:19 The "chicken and egg" problem of data flywheels
01:03:21 Embodied data: the hard part is defining the demand
01:05:18 Teleoperation, simulation, UMI, and human data
🟢 Only Optimists Can Succeed
01:09:56 Research taste in PhD interviews
01:14:08 Doing a PhD with a young assistant professor — is it a startup?
01:17:25 AI paper flood, how to read them now?
01:19:22 The embodied marathon, what kilometer are we at?
Welcome to subscribe to Crossing: 🚦 We track the industry shifts and new entrepreneurial opportunities brought by the new wave of AI technology.
🚦 Crossing is Steve Jobs's metaphor for Apple — standing at the intersection of technology and liberal arts, where great products are born. AI is transforming every industry. We seek out, interview, and bring together a new generation of AI founders and active builders in the AI era. Together with them, we explore and embrace the new changes, the new possibilities.
👦🏻 Host Koji: I founded Crossing, launched AI Hacker House — a community space for a new generation of AI founders — and serve as Venture Partner at ZhenFund. I believe technology, especially AI, represents the greatest value-creation opportunity of our generation. Koji on Jike, Koji's website
View episode transcript on Xiaoyuzhou