Kate Stephens

Software Engineer 👾 Mathemagician 🪄 Machine Learning @NVIDIA 💚 all views and opinions are my own

[Arian van Putten @ProgrammerDude]

cat is the most misused thing by programmers new to linux. I cringe every time someone uses it wrong in a bash script. Thread below with proper uses of cat only

[OpenAI @OpenAI]

OpenAI announces leadership transition https://t.co/fFYDLwGXQz

[Jim Fan @DrJimFan]

Can GPT-4 teach a robot hand to do pen spinning tricks better than you do?

I'm excited to announce Eureka, an open-ended agent that designs reward functions for robot dexterity at super-human level. It’s like Voyager in the space of a physics simulator API!

Eureka bridges the gap between high-level reasoning (coding) and low-level motor control. It is a “hybrid-gradient architecture”: a black box, inference-only LLM instructs a white box, learnable neural network. The outer loop runs GPT-4 to refine the reward function (gradient-free), while the inner loop runs reinforcement learning to train a robot controller (gradient-based).

We are able to scale up Eureka thanks to IsaacGym, a GPU-accelerated physics simulator that speeds up reality by 1000x. On a benchmark suite of 29 tasks across 10 robots, Eureka rewards outperform expert human-written ones on 83% of the tasks by 52% improvement margin on average. We are surprised that Eureka is able to learn pen spinning tricks, which are very difficult even for CGI artists to animate frame by frame!

Eureka also enables a new form of in-context RLHF, which is able to incorporate a human operator’s feedback in natural language to steer and align the reward functions. It can serve as a powerful co-pilot for robot engineers to design sophisticated motor behaviors.

As usual, we open-source everything! Welcome you all to check out our video gallery and try the codebase today: https://t.co/BHiNmqPoWE
Paper: https://t.co/bdh9TYQtHm
Code: https://t.co/lqKiaM2yYJ

Deep dive with me: 🧵

[˗ˏˋmewtru´ˎ˗ @trunarla]

Friend making normal wages- “no worries bro, I’ll cover this one. You got next!”

Friend who works in tech making over 300k- “can you Venmo me $3.74 for the sip of my drink you took?”

[Kate Stephens @realkatstephens]

Github: *goes down*
Devs:

[Noah Hein @TheNoahHein]

Devs when github is down

[Kate Stephens @realkatstephens]

My toxic trait is changing playbackRate in the console for embedded videos.

[Kate Stephens @realkatstephens]

Heat v Celtics game last night got me like

[Kate Stephens @realkatstephens]

I spend a lot of time advocating for open source and explaining the “why open source” question from non-technical folks. For those of us who advocate for OSS, I urge you to spend the time explaining - kindly and without belittling. Address the concerns. There is so much to gain!

[Kate Stephens @realkatstephens]

@unixterminal Your role is so important though! For all of us forced to work on windows machines and those of us who need corporate capabilities on Linux machines. Thank you 🙏🏻

[Greg Brockman @gdb]

Plugins for processing a video clip, no ffmpeg wizardry required. Actual use-case from today's launch.

[Bojan Tunguz @tunguz]

Kinda crazy how much the future of the World is dependent on ultraviolet lithography, and almost no one has ever heard of it.

[Cassidy @cassidoo]

🍵

[Kawin Ethayarajh @ethayarajh]

📢 Models like #ChatGPT are trained on tons of human feedback. But collecting this costs $$$!

That's why we're releasing the Stanford Human Preferences Dataset (🚢SHP), a collection of 385K *naturally occurring* *collective* human preferences over text.
https://t.co/cRY1F8TjFz

[Bojan Tunguz @tunguz]

Oh you are an AI researcher? Name all possible CUDA installation errors.

[Lex Fridman @lexfridman]

Lol, well played🤣

[Santiago @svpino]

Open this if you want to see how ChatGPT built a Deep Learning model starting from nothing.

[Bojan Tunguz @tunguz]

Roses are red
Violets are blue
Nothing says 'LOVE'
Like a new GPU!