Teahose.
SIGN IN
NEW HERE — WHAT TEAHOSE DOES
We read the entire AI & tech firehose — so you don't have to.
PODPodcastsAll-In, No Priors, Acquired…
NEWNewslettersStratechery, Newcomer…
PAPPapersPhysical AI research
PHProduct Huntdaily launches
VCInvestor ScoutSequoia, a16z, Benchmark…
CLAUDE DISTILLS →
7 reads, 30 sec each — free, 6 AM ET.
+ a live graph of the companies, people & themes underneath.
HOME/PEOPLE/JIM FAN
// PERSON

Jim Fan

ROLE RESEARCHERAT NVIDIAMENTIONS 11LAST SEEN APRIL 30, 2026
// BIO

Linxi "Jim" Fan is a Director of AI and Distinguished Scientist at NVIDIA, where he co-leads Project GR00T (humanoid robotics) and the GEAR (Generalist Embodied Agent Research) Lab. He holds a Stanford Ph.D. and was OpenAI's first research intern. Fan is best known for his work on embodied AI and physical robotics, including DreamDojo, an open-source interactive world model trained on 44,000 hours of human video that enables robots to learn from synthesized simulation without hand-authored physics engines.

Discussed in
// RECENT MENTIONS
// SIGNALS
11 SIGNALS
01
mention·arXiv Physical AI·JUNE 15, 2026

We reproduce this baseline using the GR00T N1.7 implementation and initialize from the pretrained nvidia/GR00T-N1.7-3B checkpoint (Appendix E). NVIDIA's GR00T N1.7 achieves 35% average success as EgoScale — the strongest baseline, but 30 points behind T-Rex.

Source
02
mention·晚点聊 LateTalk·MAY 18, 2026

Jim Fan and Yuke Zhu's research taste and style matched mine quite well. At the time I also really wanted to collaborate with them.

Source
03
product·晚点聊 LateTalk·MAY 18, 2026

Dream Dojo is a relatively universal world model pretrain. We open-source it so that anyone with a new robot can quickly connect to our world model, fine-tune it, and use it.

Source
04
product·Sequoia AI Ascent·APRIL 30, 2026

We pre-train Eagle Scale on 21k hours of in-the-wild egocentric human data with zero robot data whatsoever and during pre-training we predict these hand joints and wrist poses. Then in action fine-tuning we collect only 50 hours of high-precision mocap data and four hours of tele-op. That's four hours of tele-op, less than 0.1% of our training mix.

Source
05
mention·Sequoia AI Ascent·APRIL 30, 2026

UMI is perhaps one of the greatest papers ever written in robotics data and it spawned two unicorn startups. On the left-hand side is Journalists improving this design so you can wear the gripper here and on the right-hand side, Sunday made these three-finger data gloves.

Source
06
mention·Sequoia AI Ascent·APRIL 30, 2026

Anyone driving Tesla or Waymo here? Anyone? You know, when you're driving, you're actually contributing to the biggest physical data flywheel.

Source
07
mention·Sequoia AI Ascent·APRIL 30, 2026

No one can describe what happened next better than Ilya himself. If you believe in deep learning, deep learning will believe in you.

Source
08
mention·Sequoia AI Ascent·APRIL 30, 2026

Can you spot another? That's Andre right there. So Andre, we're going to the Computer History Museum.

Source
09
product·Sequoia AI Ascent·APRIL 30, 2026

These are some clips from Veo 3. You can see that the models, they pick up gravity, buoyancy, lighting, reflection, refraction, all by themselves. None of this is coded in.

Source
10
mention·Sequoia AI Ascent·APRIL 30, 2026

The last three years were dominated by VLAs or vision language action models, and models like Pi and Groot fall in this category... We call this new type of model World Action Models, or WAM. So let's all take a moment of silence for our dear friend VLAs. They've served us well, rest in peace, long-lived World Action Models.

Source
11
mention·Sequoia AI Ascent·APRIL 30, 2026

These are some clips from Veo 3. You can see that the models, they pick up gravity, buoyancy, lighting, reflection, refraction, all by themselves. None of this is coded in.

Source

AI-extracted from podcast / newsletter / paper summaries. May contain errors.