Zhejiang University

Medical staff then performed an MRI on his head and found this.

Another report from the First Affiliated Hospital of Zhejiang University tells the tale of a 46-year-old construction worker, nicknamed Chu by the Chinese researchers to protect his identity, who arrived suffering seizures and headaches.

He was taken to the hospital after his co-workers found him collapsed on the job during a seizure, and told medical staff that each night for about a month he had been experiencing seizures as if he had epilepsy.

Doctors quickly ordered up a CT scan, where they found multiple intracranial calcifications, calcium buildup in brain tissue, and lesions on the inside of his skull.

It's one of the epicenters of AI innovation for China.

Uh, Zhejiang University is there, which has got ranked number one in the world in, in research, actually ahead of Harvard for the first time in 20 years.

And so we cannot take what's- lightly what China's doing right now in this AI race.

It's very important that we continue to lead.

So what is driving this new way of exchange? How are Chinese and American creators influencing each other? Is this a short-term fad or the beginning of something bigger? Welcome to a special edition of The Point with me, Lu Xin, coming to you from Beijing.

I'm pleased to be joined on the set by Zhao Yupei, Vice Dean of the College of Media and International Culture at Zhejiang University, and Scotty Bobcox, an American actor who has appeared in more than 100 film and television productions here in China.

I'm also pleased to be joined online from Brazil by Richard Pena, Director Emeritus of the New York Film Festival and Professor Emeritus of Film and Media Studies at Columbia University.

The warmest welcome to all of you.

But we are human beings, so we can only do what we can at the best.

Anyway, it's been a very interesting and informing conversation.

Thank you so much, Yupei from Zhejiang University.

Scotty Bob Cox, an American actor who has been playing in a lot of Chinese movies and TV production.

Anyways, from there, we went sort of northwest over to Deep Robotics, their headquarters in what's called the Zijin Plaza.

It's basically just kind of north of Alibaba and really close to Zhejiang University.

And the two founders of this company are both professors from there.

So that's pretty common for these robot companies.

they're huge they're the byd people don't know about zeker spent some time with them went over to deep robotics which i've talked about before i think they're really a cool company and then sort of three days on alibaba campus bouncing around there there's a lot to do there in that we actually visited a couple companies around honjo honjo is a really interesting place because it's You know, it's known for Alibaba and what that has sort of built around it.

There's Zhejiang University, which is there.

There's some AI companies there.

DeepSeek is there.

But yeah, but building a toy, I think the idea came from many, many years ago when I was talking to John, who is another co-founder of the company.

We were university classmates and we graduated from Zhejiang University in China and we both were software engineers.

um but he quit software software engineer industry at a very early early age then he started to build his own toy um which was a bear and not a bull but a bear yeah Yeah, it was, I guess, eight, nine years ago.

And he was doing that bear character for many, many years, but couldn't sell quite well.

She was a former postdoc.

Now she's a professor at Zhejiang University.

So what she was doing was she did a forward genetic screen And she was, so in Adrian's lab, sepal is an organ of interest because the sepals actually, we've got four sepals for each flower and these four sepals have very similar shape and size and it's consistent across any flower that you look at within the same plant.

So as Adrian calls it, the sepal morphology is very robust.

Right.

And the most interesting paper in this camp is Metapoint, from ByteDanceSeed and Zhejiang University, published in June.

Their insight is really elegant.

The reason multimodal LLMs can't handle coordinates is that coordinates are continuous numbers and the model's native representational space is tokens, discrete chunks of text.