President & Co-Founder @OpenAI
Joined July 2010
- Tweets 4,815
- Following 65
- Followers 498,999
- Likes 1,386
Went to sleep at 3:30a last night trying to track down a nasty CUDA illegal memory access bug, woke up early to keep digging, & just now pushed a fix. Hard to beat the endorphin rush of doing battle with a tricky bug and coming out on top.
29
22
7
949
Step towards a general computer-using agent: openai.com/blog/vpt/
AI learns to use keyboard & mouse from video (& small amount of contractor data). With a bit of reinforcement learning, can do a task that takes proficient humans 20 minutes—crafting diamond tools in Minecraft.
3
18
3
114
Really cool to see how many people want to use cutting-edge AI tools. Seems like we're just crossing the threshold of usefulness — and we are still just at the beginning of how amazing these technologies can become:
little openai update:
gpt-3, github copilot, and dall-e each have more than 1 million signups!
took gpt-3 ~24 months to get there, copilot i think around 6 months, and dall-e only 2.5 months.
12
14
188
Big challenge with machine learning engineering is that parts of your program output lists of numbers, and the correct values for those numbers often are non-obvious. But when there's already a reference implementation, you have something to check against—goes from scary to fun.
4
7
64
Very cool to see what a creative group of kids can express with DALL·E 2: youtube.com/watch?v=y9y22D7z…
8
16
5
86
GPT-3 is now a published poet! newyorker.com/culture/cultur…
6
14
5
115
The waitlist is over; GitHub Copilot is now generally available! One of our favorite usages of Codex:
GitHub Copilot helps you get better focus and build faster by instantly suggesting code—and is now available for developers everywhere.
github.blog/2022-06-21-githu…
11
36
5
269
Always wanted to be on the cover of Cosmo:
PSA: The world’s smartest artificial intelligence (aka @OpenAI) made this magazine cover (yes, really!)—its first EVER.
7
6
1
149
Evolving the code of a simulated robot via mutations provided by a large language model:
“Evolution through Large Models” – new paper from our team at OpenAI. Step towards evolutionary algos that continually invent and improve at inventing: Large models can suggest (+ improve at making) meaningful mutations to code. Paper: arxiv.org/abs/2206.08896 1/4
2
5
64
What I find most incredible is that the idea of the artificial neural network was designed in the 1940s, and the field since then has been realizing the vision while yielding increasingly stunning results.
Feels like a collective project of humanity for 80 years and counting.
i find it both obvious and incredible that a neural network is a digital brain that lives inside a computer (and that actually kinda works)
9
26
2
207
Greg Brockman retweeted
i find it both obvious and incredible that a neural network is a digital brain that lives inside a computer (and that actually kinda works)
501
758
73
16,537
AI alignment will require talent from many fields; there are many hard open problems but we're starting to make rapid progress. Welcome, Scott!
Really looking forward to working with the legendary Scott Aaronson!
scottaaronson.blog/?p=6484
4
63
Explaining current events is fitting the train set. Predicting what comes next is generalizing to the test set. The former sees most the activity for both humans and models, but the latter is where the value is.
3
7
1
57
One of my favorite design patterns is the middleware stack—providing a simple API that is meant to be “wrapped” by a new object implementing the same API. Makes it very easy to add&reuse functionality with minimal coupling between components. Equally useful in web servers and ML.
4
13
90
While weightlighting, I've noticed my form gets better the heavier the weight. Seems like a good analogy for how startups can succeed against long odds — only when the challenge is sufficiently hard will performance rise to the occasion.
12
12
1
123
Extremely impressed by the progress the frontend ecosystem has made in the past decade. Eg writing a real-time updating app is now dead simple with React and GraphQL — in 2010, it seemed like every company was inventing their own framework. Also npm, ES, etc have come a long way.
10
7
1
145
Progress on AI alignment: we've created an AI system which can write a critique of a short story summary, successfully helping humans find hidden flaws (a task that takes humans ~10 minutes). Step towards our long-term alignment roadmap:
openai.com/blog/critiques/
3
7
2
89