President & Co-Founder @OpenAI
Joined July 2010
- Tweets 4,815
- Following 65
- Followers 495,334
- Likes 1,386
Big takeaway from the GPT paradigm is that the world of text is a far more complete description of the human experience than almost anyone anticipated.
223
419
136
3,835
Writing code for is not very fun for its own sake. What makes it insanely addictive is the feeling upon shipping — that actual people are doing something useful with what started as a figment of your imagination. Never gets old.
79
225
36
2,319
Software complexity comes in two forms — fundamental & accidental. The task of a software engineer is to discover & write down the fundamental complexity of a solution to a problem, while inventing the minimal amount of accidental complexity.
20
99
9
619
Hardest part about designing good software is pricing in potential future needs. Easy to optimize a piece of code while making it impossible to change the spec, or design clean abstractions that will be circumvented once you add a new concept. Anticipating possible futures is key
32
48
5
600
Zoom plugin to create personalized background images using DALL-E — very high usefulness to complexity ratio.
My first @ProductHunt launch is scheduled for Zoomscape AI. 🎉
Built in < 20 working hours using AI from start to finish using @OpenAI and @bubble.
9
24
2
187
Predicting price/make/model/etc from a photo of a car. Pretty cool that these days you can do this kind of thing as a side project.
Happy to share a cool hobby project I've been working on for the past few weeks: predicting attributes of cars from images.
Blog post: blog.aqnichol.com/2022/12/31…
Github: github.com/unixpickle/car-da…
Gradio demo: huggingface.co/spaces/unixpi…
7
15
1
218
Prediction: 2023 will make 2022 look like a sleepy year for AI advancement & adoption.
660
1,812
345
17,550
When editing an ML codebase, I recommend constantly taking small detours to make quality-of-life tweaks as you notice them, rather than send them off to languish in a bug tracker. Highest order bit in ML in iteration speed, and these small pains add up faster than you know.
18
26
7
453
A management lesson we learned early on in Dota.
We'd initially set goals of the form "beating X player by Y date", but when we fell behind, wasn't clear to anyone how to speed it up since we didn't know what would work.
Great results from instead "try X experiment by Y date".
13
62
4
688
Copilot, but for your spreadsheet:
I spent the last 2 weeks building the ultimate AI assistant for spreadsheets.
It can:
- Autofill entire sheets just based off context
- Generate formulas and filter rows using text commands
- Directly integrate GPT-3 into your sheets
And more…
Here are a few examples:
(1/8)
15
56
4
588
We didn't invent the idea though! First time I saw magic thread titles it was in @ludwig's @QuillChat.
10
2
75
An undervalued principle of product design is that little delights add a surprising amount of value:
ChatGPT came up with a cool UX enhancement. AI automatically names the thread for you (and its good). This should be in all AI-first products -- subject lines, Notion page titles and icons, thread names, etc.
Design principle: don't make me think.
19
41
2
543
Interesting first results of applying GPT embeddings to early diagnosis of dementia — 80% accuracy on a challenge dataset of speech recordings. Note also the paper was submitted in August; we've released much better embeddings since then! eurekalert.org/news-releases…
32
210
37
1,389
Definitely could be interesting! It won't help with the "preloading a large dataset"-use case, if your dataset takes up most of RAM. But could make for a nice server-free and terminal-magic free option if you want to load some dependencies or (maybe?) something language-agnostic.
1
5
Recently dusted off Pyseidon, a simple-to-use tool that speeds up any Python workflow when iterating with slow dependencies: github.com/gdb/pyseidon. Loving it!
Works by preloading anything you'd like in a server process, and then forking a subprocess to run the remaining code.
7
14
143
Going to be really important that we can each get AI to do what we individually want (within the broad bounds of what society deems acceptable) as AI becomes part of daily life. Will take iteration & research to get it right, hoping to share initial ideas soon.
unintended; going to take us some time to get all of this right (and it still requires more research).
generally speaking, within very wide bounds we want to enable people get the behavior they want when using AI. will talk more about it in january!
27
16
6
227
Clever integration making DALL-E dead-simple to use — a CDN which lets you generate new images just by requesting an appropriate URL: bunny.net/blog/introducing-b…
13
52
2
415