Zhen retweeted
Thanks for coming Chris! It’s always fun sharing the joy of watching Taku work live :)
Had a lot of fun briefly visiting @JasonMa2020 at @DynaRobotics on the way back to the airport. Super impressed and got to meet Taku (it works! Its real!)
Glad you guys were finally able to meet and have a reunion :)
Had a lot of fun briefly visiting @JasonMa2020 at @DynaRobotics on the way back to the airport. Super impressed and got to meet Taku (it works! Its real!)
The founders actually to invest in are the ones who get invited to but don’t go to the VC hosted events…
Good - align in November and bring it home for the G20 in December in Miami.
Good to see progress through diplomacy and ambassadorship.
China 'Already Preparing' Trilateral Summit With US & Russia, Likely For November: Kremlin zerohedge.com/geopolitical/c…
Modern technology is built by global teams, global supply chains, and shared scientific literature.
Semiconductors, robotics, and clean energy all advanced faster because researchers in different countries built on one another's work, published openly, and competed to improve shared ideas.
The strongest future for technology is a cooperative one, in which the United States and China treat scientific and technological progress as a shared project as well as a national one. Partnership does not remove competition. It gives competition a constructive purpose.
Progress grows when knowledge is shared.
I know what I’m getting my kids for Xmas
When @jensenhuang asks for a somersault, you deliver. 🦆
See how AI agents on NVIDIA RTX Spark help bring @huggingface's Microduck from simulation to the real world.
Love that all the robotics data startups are starting to come up with leaderboards, but it’s also just the 1st inning, and currently pretty one dimensional
Looking forward to the day we move past table top YAMs and get to the true frontier which is already on long horizon physical agent tasks.
Folks outside of OAI, Ant, Gemini are already doing far more advanced private evals in-house and in addition to success rate also focus on the whole system including system 0, system 1, and system 2 performance.
You’re almost certainly totally wrong 😜
Volume/commodity ego data is indeed saturated. *high quality, high diversity, high skill, good* ego data is not and the post-training and eval angle is already blossoming.
I may be totally wrong, but I'm relatively bearish on physical AI data right now. VLM data (ego) is straight forward and saturated with vendors. Prices are low. There isn't yet a clear "post training" angle where RL Labs can add a lot of value. It's early, but I think it develops slower than people expect.
A helpful example of hard tech quietly enabling consumer AI:
Cooling has traditionally required fans, which do not fit well in AI glasses or thin smartphones. Solid-state micro-cooling replaces the fan with a semiconductor device with no moving parts, providing silent, targeted cooling in a very small form factor.
xMEMS has announced that its solid-state cooling chips have entered mass production, with first commercial products expected in early 2027.
Many limits in consumer technology are thermal limits in disguise. Solving heat expands what designers can build next.
What does "AI-native" actually mean?
An AI-native application is not an existing app with a chatbot added to it. It is an application designed around AI from the beginning, where intelligence shapes the core experience: how it personalizes, how it responds, and what actions it can take for the user.
A useful test is simple. If you removed the AI, would the product still basically work? If yes, the AI was a feature. If the product would no longer make sense, it is AI-native.
That distinction will matter more, not less, as consumer applications mature.
Bryan Johnson building the market for sauna gear
one day when I get my own sauna I’m just gonna go in my birthday suit










