President & Co-Founder @OpenAI
Joined July 2010
- Tweets 4,815
- Following 65
- Followers 498,999
- Likes 1,386
One of the least-appreciated skills in programming is writing anti-frustrating error messages.
A good error message should make it self-evident (a) what the user did, (b) what acceptable inputs are, and (c) how to fix the problem.
Can determine love or hate for your library.
245
663
50
8,978
GPT-3 for fixing OCR errors:
A rare & unknown gem of a paper — academia.edu/67825612/A_Soci… (from 1996!) analyzes the root cause of the AI winters.
The story you always hear is that the neural net people overhyped everything; the real story seems to be a dedicated smear campaign by eminent scientists.
12
34
5
330
A rare & unknown gem of a paper — academia.edu/67825612/A_Soci… (from 1996!) analyzes the root cause of the AI winters.
The story you always hear is that the neural net people overhyped everything; the real story seems to be a dedicated smear campaign by eminent scientists.
13
35
4
172
One of my biggest growth moments as a programmer was realizing that libraries I use are just code, and I could read them directly rather than puzzling it out from the docs.
Even today, I am surprised how much faster I move every time I start reading a layer I'm building on.
70
273
49
3,392
"An academic is someone who has given up earthly rewards in exchange for the admiration of their peers" — @ilyasut
12
18
1
242
Shortly after moving to Silicon Valley, I remember thinking that the “tech industry” was a misnomer—only route to a startup was applying already-created technologies.
No longer—from AI to fusion to crypto, the future has been (slowly!) shifting back to inventing high technology.
4
9
120
Alignment is the challenge of designing technical & non-technical solutions to ensure that increasingly powerful AI systems further the hopes of their operator.
Probably unlike any prior technological experience, but with parallels to life experience.
3
12
2
71
AI supercomputing — the art of coaxing mind-boggling numbers of machines to perform their part of a single precisely-orchestrated computation — presents both the hardest & most rewarding technical challenge I've ever worked on.
18
30
3
479
Surprising commonality between AI & startups: newcomers tend to believe that ideas are the hard part and execution is commoditized.
In reality, many (not all!) great ideas are already in the air, and proving their value requires a rare caliber of execution.
18
65
7
521
Math competitions like AMC, AIME, and the USAMO were formative parts of my middle- and high-school experience.
I never quite made it to the IMO, but our new neural network is gearing up to be a formidable competitor at each level of the math competition circuit:
We trained a neural network that solved two problems from the International Math Olympiad. openai.com/blog/formal-math
4
26
2
315
Our API is now powered by InstructGPT, language models trained to follow instructions rather than follow the pattern set by example (which is how GPT-3 works).
Much easier to use, and much more aligned with our vision of how safe AI models should operate: openai.com/blog/instruction-…
2
26
4
183
You can now use our best models to produce embeddings (machine-readable representations) of text or code for fast & accurate search, clustering, topic modeling, and classification: openai.com/blog/introducing-…
7
75
6
524
The difference between a startup and a large corporation is that in large corporations, everyone is assigned a seat and a desk and a parking spot and a cubicle, and in startups everyone is assigned a task.
62
269
31
2,126
We've trained a GPT-3 variant which can browse the web, allowing it to give more truthful answers (& cite its sources!): openai.com/blog/improving-fa…
21
102
14
634
You can now fine-tune GPT-3 on your own data. This vastly increases the accuracy of GPT-3, making it reliable for a much wider set of tasks: openai.com/blog/customized-g…
8
50
8
326