AI, EVs, Robotics, Education, China. Mom. Also I help edit @techbuzzchina. Views personal. Ask me anything
San Francisco, CA
Joined January 2008
- Tweets2.8K
- Following918
- Followers80.7K
- Likes7.8K
Pinned Tweet
A lot more people are organizing China tech trips these days, which I actually think is great. More people should come see what’s happening here firsthand.
But here’s what I think makes @techbuzzchina different: deep domain expertise.
Our trips didn’t start as tours. They grew out of our work with fundamental investors who needed to meet companies face-to-face, ask detailed questions, pressure-test what they were hearing, and understand the environments in which these companies actually operate.
Today we work with a much broader group: VCs, founders, executives, and others across tech and investing. So the trips have expanded too. We might visit a factory, an R&D facility, or a robotics lab alongside closed-door company meetings. Every trip is built around what the group actually needs to learn.
But the underlying model hasn’t changed.
We’re not general business, marketing, or China consultants who moved into tech because AI, robotics, semiconductors, and other deep-tech sectors happen to be where the action is now.
I’ve worked in tech, finance, and investing for more than 20 years. I know the companies, the technologies, the investors, and the ecosystem. And because I live and work in Silicon Valley as well, I understand what people coming from overseas already know, what they think they know, and where the real knowledge gaps are.
That means we’re not just arranging access. We know who is worth meeting, what is worth seeing, what questions to ask, and how to put what you hear into context.
That’s the product.
More info below:
I guess we should take comfort in the fact that the robots aren't gazing at the stars yet? That's still something uniquely human? lol
Lol, just read this in Chinese:
“The world has moved so fast these past two years. We all assumed AI and robots would be washing our dishes, mopping our floors, and scrubbing our toilets, so we'd be free to write poetry, paint, and gaze at the stars.
Instead, it's the other way around — AI is busy writing the poetry and making the art, and I'm still doing the dishes and mopping the floors myself.”
I alreadt greatly prefer watching YouTube videos to trying to read inscrutable user manuals, and now it will become even easier!
Chinese manufacturers need to do this immediately. So many of the negative comments on Amazon are about poor manuals, and now here is a super elegant and fast solution.
IKEA manuals, but they come to life.
I built the real BILLY instructions in 3D, so you can watch every step happen exactly as drawn. try it: sael.net/flatpack
I'm extremely embarrassed to say that about 10 days ago (and I'm not even quite sure how it happened), something wiped my Hermes agent setup, which I didn't have proper backup for. I've been spending the last two weeks fixing things intermittently.
I now have proper backup, yay! But man, I also realized just how much work I've moved to AI, and how excruciating it is when stuff is broken.
I feel like AI people have way more AI FOMO in the Bay just because missed opportunities are so visible. Non-tech folks are barely aware of what's happening.
I’ve been attending a few dinners in SF just to catch up with old friends and meet up and coming founders.
I’m usually the only non tech founder in the room so all the tech folks don’t find me competitive and open up to me.
The one thing I’ve learned: no matter how successful the founder is (8-9 figure revenues, 10 figure valuations), they still feel the same fomo everyone else feels about AI and all the ridiculous AI exits happening.
AI anxiety is everywhere.
Muse is really good: it's proactive without being creepy, thoughtful without overthinking, and it hasn't dropped things.
I actually use it quite a bit, and I find it pretty useful for household stuff, even if all I gave it is calendar access.
I noticed today that it has a new menu of things it suggests for you. It's categorized into:
• A feed of everything it detects you're interested in
• Ideas
• Goals
• A library of files you've shared with it
Clearly they're working on improving the product and making it a full-fledged assistant. They're basically betting that they can become more of a front page, like a chief of staff proactively pushing things to you. Versus the UI for Instinct which seems difficult to make super proactive without becoming very spammy.
Meta's AI companion Muse has surpassed 10 million U.S. downloads in its first 40 days, joining an exclusive group of apps that includes Threads. New data also shows Muse is retaining users at higher rates than its AI rivals.
spr.ly/6018BHpdF0
Cool video but no actual intelligence showcased and with the use case shown it's practically begging to be banned (Remember, new models need to get permits from the Department of War). I don't understand the strategy, but this is indeed what the company is good at
It's great to boss around your consumer AI agent, but then inevitably, you need some physical arms to do some things, and that is obviously the husband. So why didn't this launch with the ability to delegate to husband from the get-go? 😆
If you asked me a few months ago, I would have been like, "Wow, these people take themselves way too seriously and are way too dramatic."
But after all the shenanigans that Anthropic has been pulling the last couple of months, I am starting to agree this might be more existential than I thought.
I'm listening to the Sold a Story podcast season two, which is on math education in the US, and it is enough to give me an aneurysm, seriously. The season opens with a bunch of little kids' voices doing very basic algebra, and you learn later that these are poor Black four-year-olds, and all of them are getting the problems correct. Oh, and they're super engaged.
Apparently, there was a group that figured out you can have kids go very fast in math if you just did direct instruction. But somehow, culturally, this was deemed "too hard" and too much of a "pressure cooker" for little kids.
It's pretty funny because it's not like there was any high-stakes testing or anything. It literally was just teaching the kids instead of "letting them discover it themselves."
I did a master's in education, and there was a lot of content in the program about equity, and how socioeconomic class, parental habits and attitudes, and level of education really shape a kid's academic achievement. It's just infuriating to hear that when they discovered a teaching method that works in spite of these disadvantages, it was thrown out as being, I guess, somehow not "fun" enough. It's difficult to believe that they didn't actually have ulterior motives.
I mean, yeah, duh. I would have been shocked if this didn't happen.
Exclusive: DeepSeek is helping Huawei design its next generation of AI chips, with Huawei engineers working inside DeepSeek’s offices.
The collaboration could help China train advanced AI models without relying on Nvidia.
Read more: thein.fo/4jqD8cr
Rui Ma retweeted
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
Another day, another story I hear of a small consulting or finance firm firing all their employees and replacing them with AI
gg junior positions
I feel like every parent I talk to kind of half-jokes about this: "Oh, what's the point of school?" And yet, they send their kids straight into the traditional schooling system anyways, one they know hasn't adequately prepared the youth of 2026
So I don't understand how it will magically fix itself to prepare the youth of 2036
LOL, this post tells you about a great business opportunity: take a $7 smart ring (like an Oura) from Pinduoduo, etch a cross on it, and sell it to religious folks for $249.
I actually went and looked up these super cheap smart rings. They really do exist at this price and are supposedly accurate, but poorly made. They probably don't hold a charge very long and may be uncomfortable to wear, but even at this price, their sensors apparently are accurate lol. And when you "pay up" in the $30-$50 range, then you're getting something quite decent. The software used to be terrible, but I wonder if that problem will be solved very quickly with the proliferation of AI coding
说个暴论：想发财，就去卖信仰！！！
拼多多上7美元一枚的戒指，刻个十字架，改名叫「祈祷提醒戒」，标价249美元。
就这么一个东西，一个月卖出11.3万美元，差不多81万人民币。
你在拼价格、拼流量，人家在卖心里的那点念想。
同样一件小东西，你卖的是商品，他卖的是信仰，价格能差50倍。
照着做只要五步：
1、找货：挑一件10美元以下的小东西
2、加印记：在上面放一个十字架
3、改名字：用信仰里的话重新起名
4、找对人：在Facebook上卖给信基督教的女性
5、等结果：看看会发生什么
美国有2.3亿基督徒，线上几乎没人专门卖东西给他们。
因为会在网上赚钱的人，很少会去教堂。
这块空白，就是你没看到的钱。
This is why I think you need to use an app extensively before commenting on it.
As for flow, WeChat is the only app I’ve ever managed to get people over 80, with otherwise no computer skills, to use. Every single one learned quickly. Trying to get them to send a regular text? Good luck.
WeChat has so many small optimizations, even if you only use it for messaging. For example, you can recall a message within what I believe is two minutes of sending it. If it’s a text message, you then have the option to edit and resend it without retyping everything. Most of the time, I’m recalling a message because I made a typo, so that saves me a step. That’s a recent change btw and just one example of the thought that goes into continually upgrading the experience.
I do get frustrated with many of the mini apps, but they’re mini apps. I don’t have to download a full app or wait for software updates, so I have different expectations. They're effectively somewhere between a mobile web page and a full-on mobile app. And when's the last time you loved interacting with a mobile web page? Not very often, I think.
And WeChat is basically the entire (Chinese) internet, with so much contained within it. At this point, a bigger problem is that, for many things, the most current information is only on WeChat. You can’t reliably find it on the open web anymore. The search has also improved a lot over the past few years, and you can search through literally anything, including all your friends' moments. Have you tried a Facebook search recently to see what you can find?
I recently gave someone the wrong opening hours for the Great Wall because I Googled them instead of checking the relevant WeChat official account.
Regarding China’s newly proposed restrictions on exporting sensitive technologies, including certain frontier AI models, I think I saw more people freaking out on X than in China. But then, of course, the algorithm is probably more likely to feed me those people.
If you understand China’s overall strategy, I don’t think you should interpret this as an attempt to kill open source. What China wants is greater control over strategically important technologies while encouraging their adoption around the world. Those two goals aren’t contradictory.
The government has made it pretty clear that it wants Chinese AI products and services to expand internationally. API access, SaaS, and lightweight on-device deployments are explicitly excluded from the proposed foundation-model export controls. Even for certain restricted technologies, including new open-source releases, it’s proposing simplified licensing procedures that could make exporting relatively straightforward.
Where things get complicated is with the core technology itself, including potentially model weights, training techniques, and transfers of intellectual property or ownership. Certain frontier models could face additional scrutiny based on their size, training compute, or performance. Exactly how these rules will apply to new open-weight releases remains an important question, although I personally expect more scrutiny of the underlying know-how than of the weights themselves.
My bigger hypothesis is that China is establishing this regulatory framework before it really needs to use it aggressively, at least when it comes to AI. It’s setting up the ability to exercise greater control in the future, when Chinese AI may be in a much stronger competitive position.
Right now, China hasn’t established a decisive overall lead in frontier AI, so it has strong incentives to keep its ecosystem relatively open. That helps Chinese developers catch up, accelerate innovation, and encourage global adoption. And China isn’t promoting open source simply out of generosity. The more widely Chinese models are used, the more commercial opportunities and influence Chinese companies can build around them.
That’s why I’d be surprised if the eventual rules were designed to broadly discourage Chinese companies from open-sourcing their models. Being overly restrictive would undermine much of what China is trying to accomplish.
Of course, there are legitimate national-security concerns, particularly around AI technologies that could enable cyberattacks or threaten critical infrastructure. China wants its AI capabilities to spread globally, but that doesn’t mean it wants unrestricted transfers of every underlying technology.
And frankly, allowing dangerous AI capabilities to proliferate without safeguards would be no bueno for world peace, or world stability.
Interesting post on China’s manufacturing advantage at the firm level, and perhaps even at the individual technician level. But the broader ecosystem of specialized and integrated manufacturing clusters is especially hard to replicate
As I’ve mentioned many times, this system also comes with real trade-offs. Once the local govt gets permission to invest heavily in a sector, they're effectively committed for at least a decade. You have to make that supply chain work: the costs are already sunk, and the resources tied up in it leave less room to pursue something else
Agree that automation and learning loops are going to be very helpful here, but China is moving quickly on this too
I've spent years in China building robots. I can tell you this playbook won't work:
Pre-seed $: Invent the ultimate actuator.
Seed $$: Figure out mass production.
The hard part isn't inventing a beautiful topology. It's process engineering.
The real delta between the US and China is the ability to execute the enormous amount of physical x mental work required to iterate thousands of times on product designs and manufacturing processes. Partly because much of the tacit knowledge / management systems has left the US, but also because of the sheer human labor required.
This isn't something you can solve by buying traditional automation lines either. It requires:
1. Ideation driven by human intuition. Walking up and down the line, talking to assembly workers, and developing hypotheses about what and where improvements can be made. Raw human hours and attention spent understanding a specific manufacturing process.
2. Large batches of experiments across multiple variables. Coordinating experiments that can easily involve 50+ people, issuing instructions, and tracking the execution and results of each experiment.
3. Purpose-built test design to identify key metrics. For example, a change in gear hobbing profile might require accelerated aging tests on a dyno, while a change in bearing tolerance might require NVH testing.
4. Data + human intuition-driven analysis. Interpreting test results and deciding whether to implement a process change or run the next set of experiments.
5. Strict process control. Reissuing work instructions to the line and ensuring every process is followed to the letter.
The best manufacturers in Asia, from Foxconn to TSMC, have built their competitive advantage on systematically turning thousands of physical experiments into better manufacturing processes.
Their moat isn't just production scale. It's the organizational machinery that makes manufacturing learn and improve faster than anyone else.
I don't think most people in SF clock this.
Say you're manufacturing actuator gears in-house in the US. After you've made your first article, will you hire a team of 10 process engineers to coordinate the 50 technicians running concurrent experiments with designs, processes, and materials?
To understand wear patterns across every pin in a cycloidal drive after 10,000 hours? To reduce backlash from 15 arcmin to 5, then to 1? Just to keep pace with China's progress over the next few years?
Do you have the skill set and management systems in place to do this?
That's why our approach isn't just to automate manufacturing. It's to automate the process of learning how to manufacture better.
Automate simulation and design optimization.
Automate assembly and physical testing.
Automate the design and execution of process experiments.
Feed real-world experimental data back into simulation to continuously improve sim-to-real accuracy.
The goal is to turn what currently takes 50 engineers coordinating thousands of experiments into a closed-loop system that continuously learns and improves.
Then the same robotic line mass-manufactures the product.
PS: Coming back to actuator design, many actuator designs are already commoditized in China. The project in the video is open source, and CNC machining cost in China is just $27 (watch the video).
I hope people understand this reality. I'll link more examples in the replies.