President & Co-Founder @OpenAI
Joined July 2010
- Tweets 4,815
- Following 65
- Followers 495,334
- Likes 1,386
A new tool for scriptwriters:
Introducing Dramatron, a new tool for writers to co-write theatre and film scripts with a language model. 🎭
Dramatron can interactively co-create new stories complete with title, characters, location descriptions and dialogue.
Try it yourself now: dpmd.ai/dramatron-github
99
369
24
3,329
Good articulation. I usually call this the “demo gap”—the distance between what looks like it’s working in a demo and what really works in practice. A deep issue that applies to humans too—eg someone who interviews well but can’t do the job. We are closing the gap but much to do.
My one-sentence summary of the AI right now is that it does amazing stuff but it has a high error rate.
This is why it both impresses and frustrates: the times it does really well go viral on twitter, the times it makes a mistake frustrate people using it as a work aid.
4
6
1
143
Steady progress by the team on Whisper. Particularly improved for non-English speech transcription. Give it a try!
Whisper large-v2 from @OpenAI, a state of the art model for speech recognition 🏆
Same data, more training => transcribe speech with 5-10% improved accuracy 🚀
Available now in 🤗 Transformers
huggingface.co/openai/whispe…
7
17
5
149
Best part about ChatGPT development & deployment is that AI alignment has graduated from being the responsibility of a single team within OpenAI — it's now a part of the practical work of many technical teams within the company.
10
23
2
299
Few things more fun than building & deploying AI systems at ever increasing speed, scale, and alignment.
19
16
4
325
We also (amazingly!) appear to be on trajectory for this part of the post: "People who have no experience in a field will suddenly be enabled to operate in that field alone. Meanwhile, people who already had experience in that field will now be even more effective and efficient."
8
16
2
158
A decent prediction on HN (news.ycombinator.com/item?id…): "Well, right now we're in the "flip-phone" era of AI. In 10 years from now, mark this prediction: nearly everyone will have an AI counterpart that is trained and unique to each individual."
19
86
16
543
Microsoft partnership has been one of OpenAI's secrets to success — we work very closely with Azure to produce AI training & serving infrastructure that can scale to our cutting-edge (& entirely unprecedented!) needs.
microsoft, and particularly azure, don’t get nearly enough credit for the stuff openai launches. they do an amazing amount of work to make it happen; we are deeply grateful for the partnership. 🙏 they have built by far the best AI infra out there.
28
151
11
1,743
Exciting but overlooked that ChatGPT is primarily an alignment advance—the base model (GPT-3.5) has been available in publicly for many months, but making it into a useful chat system required significant strides with reliably following the intent of the developer and the user.
iterative deployment is, imo, the only safe path and the only way for people, society, and institutions to have time to update and internalize what this all means.
49
126
30
1,300
Just launched ChatGPT, our new AI system which is optimized for dialogue: openai.com/blog/chatgpt/.
Try it out here: chat.openai.com/
78
358
75
2,328
My perspective on AI safety has evolved as reality has increasingly diverged from my expectations, formed from reading various AI thought experiments 10-15 years ago. A nice essay by Scott Aaronson which captures the emerging landscape: scottaaronson.blog/?p=6821
9
22
7
151
I've resisted using mosh (mosh.org/) for years because its terminal handling is not quite as good as ssh's, but as of today I'm sold on using it any time I travel. Something magical about opening your laptop and continuing your remote session like nothing happened.
6
3
90
Instant signup for DALL-E:
We’ve removed the waitlist for DALL·E. Sign up and start creating immediately! openai.com/blog/dall-e-now-a…
40
10
1
144
Whisper, a neural net that approaches human level robustness and accuracy on English speech recognition. Attached is transcriptions of the same voicemail with iOS vs Whisper. Available today as open-source: openai.com/blog/whisper/
24
113
20
806
DALL·E Outpainting — extend an existing image, to arbitrary size: openai.com/blog/dall-e-intro…
26
181
16
1,550