Tianjun Xiao

Principal Engineer @Nvidia
Before: Senior Applied Scientist | Manager @AWS, Staff Machine Learning Scientist @Tesla Autopilot

[Jensen Huang @JensenHuang]

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.

AI will transform every industry, power every company, and be built by every country.

Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

The world needs both frontier closed models and frontier open models.

https://t.co/AUKzoQ5Ikb

[Marco Pavone @drmapavone]

Jensen today announced Alpamayo 1.5 at #NVIDIAGTC!

#Alpamayo 1.5 is a major update to Alpamayo 1—@nvidia’s open 10B-parameter chain-of-thought reasoning VLA model, first introduced at #CES. Built on the #Cosmos-Reason2 VLM backbone and post-trained with RL, it adds support for navigation guidance, flexible multi-camera setups, configurable camera parameters, and user question answering. The result is an interactive, steerable reasoning engine for the AV community. We’re also releasing post-training scripts to help researchers and developers adapt the model.

Additionally, we’ve significantly expanded the Alpamayo open platform across data and simulation, including releasing highly requested reasoning labels for the PhysicalAI Autonomous Vehicles dataset (https://t.co/fD9eUcndya), as well as our chain-of-causation auto-labeling pipeline.

🔎 Learn more about Alpamayo 1.5 and the latest extensions to the Alpamayo open platform:
https://t.co/P0nuqkwBab (please note that most of the links will become active in the next few days.)

Happy building—and stay tuned for more in the coming months!

@NVIDIADRIVE @NVIDIAAI

[NVIDIA @nvidia]

What does comfortable, confident, safe autonomy feel like?

Ride through San Francisco with NVIDIA founder and CEO Jensen Huang and NVIDIA VP of Automotive Xinzhou Wu as they discuss the technology powering the next generation of autonomous vehicles.

[Tianjun Xiao @sneakerkg]

This framework work gives me a deep impression where vibe-coding is right now. You can basically see the complexity by the diagram below. @bingxu_ have you tried to attach @karpathy's nanogpt code onto your backend?

[Bing Xu @bingxu_]

Just open-sourced VibeTensor — the first deep learning system fully generated by an AI agent, with 0 lines of human-written code: https://t.co/8wwe1LB7xK

It’s a working DL system with RCU style dispatcher, a cache allocator and reverse-mode autograd. The agent also invented a Fabric Tensor system — something that doesn’t exist in any current framework.

The Vibe Kernel includes 13 kinds and 47k LOC of generated Triton and CuteDSL kernels with strong performance.

VibeTensor was generated by our 4th-generation agent. It shows a “Frankenstein Effect”: the system is correct, but some critical paths are designed in inefficient ways. As a result, performance isn’t comparable to PyTorch.

I haven’t written a single line of code since summer 2025. I started this effort after @karpathy 's podcast — I didn’t agree with his arguments, so Terry Chen and I began using it as a stress test for our agents. The “Frankenstein Effect” ended up exposing some of our agent’s limitations — but the direction is clear.

[Zhuang Liu @liuzhuang1234]

A neat result: for many video tasks, always-on CoT is redundant or even harmful. VideoAuto-R1’s reason-when-necessary “Thinking Once, Answering Twice” with confidence-based early exit is a clean way to get accuracy and efficiency.

Congrats to @YoungXiong1, Shuming & others for leading the work!

[Tianjun Xiao @sneakerkg]

@ZechenBai Right way to go

[Tianjun Xiao @sneakerkg]

Check out our latest efforts. Inference time scaling would be an interesting topic to dive into.

[NVIDIA DRIVE @NVIDIADRIVE]

Curious how reasoning-based autonomous vehicles are built in practice?

See how NVIDIA Alpamayo brings together open models, datasets, and simulation in a full reasoning-based AV workflow:

🔹 Generate trajectory predictions with reasoning traces using Alpamayo 1, available on @huggingface
🔹 Train and evaluate models with Physical AI Open Datasets
🔹 Test end-to-end performance in AlpaSim, an open-source closed-loop simulator

Get started → https://t.co/R99eztTWa7

#CES2026

[Tianjun Xiao @sneakerkg]

Wow, still remember this coast-to-coast drive was a target Elon set to the team a while ago and it came true now. Can learn a lot from the process and timespan how technology is developed.

[Ashok Elluswamy @aelluswamy]

World’s first fully autonomous coast-to-coast drive, done with Tesla self-driving v14.

Congrats and thank you @DavidMoss!

[Andrej Karpathy @karpathy]

I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There's a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.

[NVIDIA DRIVE @NVIDIADRIVE]

🚕 Robotaxis are moving from pilots to real streets.

From Austin and Dallas to Las Vegas and beyond, industry leaders explain how autonomous vehicles are scaling safely—using real-world data, simulation, and AI built for edge cases.

Hear from @Uber, @Avrideai, and @zoox on what it takes to commercialize robotaxis and raise the safety bar for urban mobility.

▶️ Watch the full video: https://t.co/HmBvtP4Bk7

[NVIDIA DRIVE @NVIDIADRIVE]

🚗 AV teams need simulation that matches the real world.

@nvidiaomniverse NuRec and Cosmos workflows turn fleet data into interactive simulation, generating weather, lighting & terrain variations for true edge-case testing.

⚒️ Neural reconstruction
🌧️ World foundation models
🎯 Scalable validation

Explore the workflow 🔗 https://t.co/pFCAbHr5yS

[Marco Pavone @drmapavone]

We’ve just released @nvidia #DRIVE Alpamayo-R1 (AR1) — the world’s first industry-scale open #reasoning #VLA model for autonomous-vehicle (AV) research. AR1 integrates Chain-of-Causation reasoning with trajectory planning to improve decision-making in complex driving scenarios.

Built on @nvidia #Cosmos #Reason, AR1 is designed as a customizable foundation for a broad range of AV applications — from instantiating an end-to-end backbone for autonomous driving to powering advanced, reasoning-based auto-labeling tools.

Resources:
Model: https://t.co/9nI9L08LJJ
Inference Code: https://t.co/QpPzLEsFnm
Paper: https://t.co/8PSdQNDqwO
Blog Post: https://t.co/S92N6ff58L

A subset of the data used to train and evaluate AR1 is available in the @nvidia Physical AI Open Datasets: https://t.co/fD9eUcndya

AR1 can be evaluated using AlpaSim (https://t.co/9Wqutgpe5d), @nvidia's newly released open-source AV simulation framework built specifically for research and development. (Separate post on AlpaSim coming soon.)

This release completes @nvidia’s trifecta — model, data, and simulator — to accelerate research and development in the autonomous-vehicle domain. Happy developing, and stay tuned for more!

Huge thanks to the phenomenal team that made this possible @NVIDIAAI @nvidia.

[NVIDIA DRIVE @NVIDIADRIVE]

NVIDIA DRIVE Alpamayo-R1 is now available on:

🔗 GitHub: https://t.co/bz0T7SWue0
🤗 Hugging Face: https://t.co/LrXDcNgPbW

[Andrej Karpathy @karpathy]

I am unreasonably excited about self-driving. It will be the first technology in many decades to visibly terraform outdoor physical spaces and way of life. Less parked cars. Less parking lots. Much greater safety for people in and out of cars. Less noise pollution. More space reclaimed for humans. Human brain cycles and attention capital freed up from “lane following” to other pursuits. Cheaper, faster, programmable delivery of physical items and goods. It won’t happen overnight but there will be the era before and the era after.

[Saining Xie @sainingxie]

you can’t build superintelligence without first building supersensing

[Akshay 🚀 @akshay_pachaar]

A team at DeepMind wrote this piece on how you must think about GPUs. Essential for AI engineers and researchers:

[Tianjun Xiao @sneakerkg]

@jiqizhixin @grok summarize this paper

[Tianjun Xiao @sneakerkg]

The necessary technical path

[Jim Fan @DrJimFan]

Evaluation is the hardest problem for physical AI systems: do you crash test cars every time you debug a new FSD build? Traditional game engine (sim 1.0) is an alternative, but it's not possible to hard-code all edge cases. A neural net-based sim 2.0 is purely programmed by data, grows more capable with data, and scales as the fleet data flywheel scales.

[Tianjun Xiao @sneakerkg]

@ZechenBai @MikeShou1 Congratulations, Zechen.

[Charles Qi @charles_rqi]

Tesla Robotaxi: A New Era Begins

I’ve (very fortunately) been part of multiple robotaxi launches. But this one is different and feels much more profound. It’s a paradigm shift. It’s the GPT moment for real-world autonomy.

Tesla’s robotaxi runs vision-only -- no lidar, no radar, no HD maps. Just video in, actions out. A single large model trained end-to-end. Many experts in autonomy doubted this could ever reach L4. Some still do. But from 10/10 We, Robot to factory driverless ops, and now a live service in Austin, Tesla keeps proving them wrong.

What made this possible isn’t more sensors or more complex engineering, but a different philosophy. We bet on end-to-end learning, powered by massive data and fleet-scale reinforcement. This is the Bitter Lesson for physical-world AI:
Hand-coded rules, curated maps, and engineered pipelines don’t scale. Data and compute do.

Now, let the scaling begin.