President & Co-Founder @OpenAI

Joined July 2010
Filter
Exclude
Time range
-
Near
A debugging technique I've been perfecting recently: when it's late at night and I've been banging my head on some bug (but have my mind fully wrapped around the symptoms), go to sleep. In the morning, usually solve it within 15 minutes. Subconscious really knows what it's doing.
83
71
24
1,143
Critics are an important part of society’s immune system against wasting resources. To be effective, their ferocity ought to be proportional to resource investment, without regard to society’s benefit from success. Means the most important endeavors will be the most criticized.
5
9
51
And as an illustration, here's "The journey begins with a single gradient step." by DALL-E. To be fair, that step does kind of have a color gradient!
3
4
43
The journey of training even the greatest language model begins with a single gradient step.
3
8
1
78
Common wisdom is that there are three types of software engineers: 1. Pioneers: build prototype 2. Settlers: scale from few to many users 3. City planners: make it rock-solid But best career impact comes from doing both 1&2 or 2&3—requires rare ability both to dream & execute.
11
59
11
461
DALL-E has a sense of style:
I asked @JackHessing1 to ask DALL·E 2 by @OpenAI to generate @BALENCIAGA 2050 Fall Ready-To-Wear. Then I sewed it now walking in real life. What do you think??? Original: labs.openai.com/s/95s2FWYwP6… Me: #ai #fashion #dalle2 #dalle
5
7
68
Also serves as the world's most advanced way to keep your kids entertained.
3
45
Novel product concepts, via kids playing with DALL-E:
OMG #dalle is the ai singularity for creativity🤯 In 5 minutes, my kids were able to create a "chicken hat", a toilet car, puppy slippers and a chicken toilet 🤣 Post your dumbest product idea below and I'll use it to turn it into something people would want to buy.
2
9
1
96
We do build on Azure, but also operate at such an unprecedented scale that we have to solve problems at essentially all layers of the stack.
2
An illustration that is all-too-applicable to this tweet — "network switch that looks like a monster rampaging a datacenter" as generated by @sidorszymon
3
3
40
Was thinking about my single most important tool for programming productivity, and had a surprising realization — it's not emacs or mypy, it's Slack. The job is about working in close coordination with a team to achieve something you couldn't on your own.
4
4
104
As an illustration – "The systems are down" by DALL-E 2. Happy that it's just the dev cluster!
3
3
47
That feeling when your phone starts blowing up with alerts, indicating everything is down... and then you realize the alerts are all for your dev cluster 😅.
2
59
Essay co-authored with GPT-3 was selected for the 2022 Best American Essays anthology:
Best American 2022 selection: Vauhini Vara (@vauhinivara)'s "Ghosts," edited by @chameauleon, in @believermag! believermag.com/ghosts/
1
25
2
107
A book of 1000 paintings of robots. Generated by a robot, curated by a human:
I spent most of this past weekend experimenting with DALL·E 2, OpenAI’s new AI system that can create realistic images from a written description. I curated a book of 1000 robot paintings. You can view the entire thing online at archive.org/details/11111010… #dalle
11
9
3
114
Unpopular opinion: don't rely on implicit truthy constructs in your language, and instead always convert to bool yourself. For example, in Python rather than "if mylist:", do "if len(mylist) > 0:". An example of trading more keystrokes for less cognitive burden for readers.
19
24
6
419
Kinda cool that any tweet can now be illustrated:
11
10
106
The mathematics of measuring how well our true objective (e.g. helpful or accurate responses) is being optimized when training an AI systems on a proxy metric (e.g. the output of a reward model). Pretty cool & not obvious that you can measure this at all! openai.com/blog/measuring-go…
1
4
1
44
Software engineering: 50% understanding requirements, 40% complexity management, 9% debugging, 1% solving "interesting" algorithmic problems. You'll enjoy software engineering a whole lot more if you instead think of the first 99% as the interesting part.
33
214
20
1,466
Big challenge in ML engineering: finding aggregate views that let you quickly understand the micro details of everything happening in your system. Often the biggest problems are simple problems obvious from looking at one specific computation. ML rewards attention to detail.
6
32
1
195