President & Co-Founder @OpenAI
Joined July 2010
- Tweets 4,815
- Following 65
- Followers 495,334
- Likes 1,386
Code is a liability, not an asset. So goal of software engineer is delivering the maximum amount of desired functionality at the cost of the least amount of code complexity, even as desired functionality evolves over time.
166
420
56
3,877
ML engineering productivity hack: write small self-contained files testing your understanding of any semantics you're fuzzy about (e.g. how CUDA streams interact with a full CUDA queue). Goal should be to turn these semantics into "muscle memory" before working end-to-end.
15
40
1
466
Very cool interface for exploring @lexfridman podcasts — can navigate through Whisper transcripts or GPT-3 summaries, on multiple timescales. Starting to feel like a new way to consume content:
16
69
7
598
Something beautiful about the purity of a tool hyperoptimized for its purpose, like how rg (github.com/BurntSushi/ripgre…) is blazingly fast to run & fast to type.
3
2
1
73
Surprised by just how much I feel understood & supported when Copilot suggests exactly the line of code I intended to write, especially while I'm still stuck mentally formulating it.
25
24
6
517
Just open-sourced Point·E — DALL·E but for 3D point clouds. Extremely fast to generate: github.com/openai/point-e
24
119
18
907
GIF
Just open-sourced our screaming-fast BPE tokenizer:
We just open sourced our tokenizer! This should help all users of our API better understand their usage. It's also incredibly fast – 3-6x faster than other open source alternatives!
github.com/openai/tiktoken
9
15
229
Training of deep learning models by “hill climbing” seems like it should be impossible because of our intuitions from 3D life. Turns out they don’t prepare us to understand what’s possible in billion-dimensional spaces.
24
36
3
522
Sometimes algorithms & data >> compute.
Just released a new embedding model which outperforms our previous best one while being 99.8% cheaper: openai.com/blog/new-and-impr…
5
6
2
129
Just released a new embedding model which outperforms our previous best one while being 99.8% cheaper: openai.com/blog/new-and-impr…
18
73
15
614
Few technical experiences are more difficult than being an Emacs user with a broken ctrl key.
22
3
1
107
Another angle on a chat interface to GPT-3.5, together with external integrations to make it even more useful:
🚀 Introducing Chatsonic by @WriteSonic
Like ChatGPT but with superpowers 🦸
- integrated Google search for up-to-date content
- image generation using @StabilityAI
- voice commands so you don't have to type
Support us on PH: producthunt.com/posts/chatso…
7
17
1
109
A surprising amount of machine learning performance engineering starts by getting fed up with how long you spend waiting for your job to launch.
13
15
1
287
Feedback greatly appreciated! Would a “power user” interface to the same model restore the magic, or does the model itself feel less fun for serious prompt engineering?
23
2
2
192
Love the community explorations of ChatGPT, from capabilities (github.com/f/awesome-chatgpt…) to limitations (cookup.ai/chatgpt/where-it-l…).
No substitute for the collective power of the internet when it comes to plumbing the uncharted depths of a new deep learning model.
30
198
16
1,093
We believe in shipping early & often, with the hope of learning how to make a really useful and reliable AI through real-world experience and feedback. Correspondingly important to realize we're not there yet — ChatGPT not yet ready to be relied on for anything important!
ChatGPT is incredibly limited, but good enough at some things to create a misleading impression of greatness.
it's a mistake to be relying on it for anything important right now. it’s a preview of progress; we have lots of work to do on robustness and truthfulness.
30
44
12
520