President & Co-Founder @OpenAI

Joined July 2010
I fell in love with programming right before college. I remember surprise & delight when I realized that college was prepared to give me credits for getting better at my newfound hobby. Even today I am amazed that it’s possible to be employed to program. The joy has not faded.
11
12
2
274
The file system makes for a great key-value store but a poor general database — writing and reading a key is very fast, but it's slow to list (with 100k+ entries), and many tools choke on giant directories. Still, sometimes can be a great decision to use it to back your app.
4
4
2
74
Modern feature that has most changed my use of Python: dataclasses. Not just useful for data structures — they let you separate state from implementation of any class. Also dataclasses tend to be more useful by default, since they come with repr, eq, etc.
6
21
2
275
First rule of system administration — a service tends towards one of three states: manually watched, monitored, or broken.
1
4
83
A software skill people not often appreciated outside security: probing the emergent semantics of a complex system. First understood this at Stripe while integrating various financial backends. Turns out to be extremely useful for building systems for training neural networks.
4
11
110
A small taste of the kinds of content that you will be able to produce personally in the not-too-distant future:
Back in 2018 at OpenAI, a few of us wrote a story with gpt as an AI "co-author". We didn't have an AI illustrator back then, but now we sort of do, so I tried plugging the text into #dalle. Here is the result! “The Bees”, a short story by humans & AIs: web.mit.edu/phillipi/www/the…
12
12
93
Any technology comes with benefits and commensurate risks. AI is unique only in the unprecedented magnitude we expect from both. Navigating these risks & distributing these benefits is daunting a priori, but each day of progress & experience makes the puzzle a bit more clear.
7
14
1
76
"A photo of an artificial neural network growing in a garden":
6
5
2
72
Tending to an ML training run is like gardening. Need to sprinkle your budding ML model with fresh data, shine energy on it in the form of compute, and nip off any unpromising growths in the form of divergences. Main difference is that plants at maturity don't speak English.
9
6
111
GPS maps made getting lost largely obsolete. Search engines made recall of specific facts far less important. Very curious what new capabilities AI systems will give us that we will soon take for granted.
45
27
5
293
Coding is pure creation.
17
28
5
294
The gradients must flow.
4
14
172
Being an ML engineer sometimes feels more like being a manager than an individual contributor. When writing software, you think through every edge case & behavior yourself. In ML engineering, you shape the behavior of systems through indirect means. Trickier, but more leverage.
16
20
2
365
I can attest that it was delicious (real cake + generated image):
4
5
34
A universal design trick in software: make as much of your application stateless as possible, and concentrate your persistent state in a central, simple, perhaps unscalable solution (database, custom process). State is a fundamentally-hard part of software, so work to minimize it
17
43
2
389
Spaghetti is meant for eating, not code.
8
9
1
61
Realized today one major difference between in-person and virtual meetings: laughter. IRL, I’ll often make jokes, but over two years of video calls I can barely recall a single funny moment. Surprised how easy it was to forget an entire dimension of interaction, and not notice.
15
22
6
217
How to succeed at a startup: get one big idea right, and then execute diligently for years on innumerable small features (e.g. being able to directly copy-paste working API code from your website) to delight your users.
10
10
1
199
The difference between a framework and a library is that a framework controls how other code is written, while a library is designed to be composed with any other kind of code. Explains why developers love writing frameworks, even though users basically always want libraries.
9
41
3
376
A surprisingly catchy (and comprehensive) song about AI progress. Will definitely be a hit at all the SF parties:
From Yann LeCun (@ylecun) to Andrew Ng (@AndrewYNg), conversational AI is full of legendary supermodels. So we created a music video honoring the history & brilliant makers behind this incredible tech, while also poking some fun at our own #AIsupermodels.
3
13
2
57