President & Co-Founder @OpenAI
Joined July 2010
- Tweets 4,815
- Following 65
- Followers 495,334
- Likes 1,386
Shows that there’s room for improvement with the next version of DALL-E.
I asked #dalle to generate paintings that AIs like but humans don’t. Nice try, AI, but I actually like SEVERAL of these paintings!
4
1
72
Very cool to see how useful Copilot already is. Easy to forget that it’s still just the beginning of the road for this technology, and in a few years it’ll be astronomically more useful than it is today.
8
8
147
Each year, we've built something that makes our previous work look antiquated—a trend across both OpenAI & the field. The tech is getting powerful, and the only way to continually ship 10x better models to the world is for industry to work together to grow into worthy custodians.
4
3
30
What makes our approach different is that we're starting with a practical, useful foundation that companies have an incentive to opt into — these best practices are grounded in hard-won experience with how these models are used & abused in practice. (7/8)
1
2
11
The best practices we developed, with cooperation & support from others in the industry, are the most concrete step so far towards building industry regulations for AI. We'll welcome thoughtful government regulation, but can't delay for it given how fast the tech is moving. (5/8)
1
1
8
Credit card processing ended up being primarily industry-regulated for historical reasons — the networks established rules with a strong eye towards consumer protection early on, and so there was not a strong public need for government to upend the industry. (4/8)
1
1
9
It didn't have to be this way. Many other financial sectors are primarily government-regulated, which makes innovation much harder & rules often end up not matching the needs of the present day. This results in more consumer protection, at the cost of less useful products. (3/8)
1
1
6
The fact that card processing is industry-regulated is very beneficial for consumers, since the rules can adapt when new paradigms appear (e.g. to allow crowdfunding—great but risky innovation, since card networks might have pay back consumers if the maker doesn't ship). (2/8)
1
1
7
One note from my experience at @stripe — surprisingly, the credit card processing industry is primarily industry-regulated. The card networks like Visa make rules, which get pushed to their member banks; those banks make rules for the processors who run on top of them. (1/8)
We're working with others in the nascent industry of creating & deploying language models in order to produce safety best practices.
Ultimate goal is to have a practice for cooperating on increasingly high-stakes safety problems as AI progresses.
6
9
72
We're working with others in the nascent industry of creating & deploying language models in order to produce safety best practices.
Ultimate goal is to have a practice for cooperating on increasingly high-stakes safety problems as AI progresses.
Preliminary best practices for language model providers, authored together with @CohereAI and @AI21Labs. openai.com/blog/best-practic…
1
4
1
50
Most surprising part about getting older is how much harder it becomes to get good sleep.
27
24
4
252
A pretty good hypothesis on how this arises (as an artifact of how the text input space is tokenized):
I took a look at the BPE encoding of the name DALL-E uses for birds.
Its "apo, plo, e</w>, ,ve, sr, re, ait, ais</w>".
Apo-didae & Plo-ceidae are families of birds, each with 100+ species. Apo-diformes is the biggest order of birds with 400+ species of birds.
3
3
49
Turns out DALL-E can read the seemingly gibberish writing it produces. Built its own mini-language that is consistent between its text input space and image output space:
DALLE-2 has a secret language.
"Apoploe vesrreaitais" means birds.
"Contarra ccetnxniams luryca tanniounons" means bugs or pests.
The prompt: "Apoploe vesrreaitais eating Contarra ccetnxniams luryca tanniounons" gives images of birds eating bugs.
A thread (1/n)🧵
8
23
9
132
A well-architected software project is one where the difficulty of adding a feature is proportional to the delta in functionality you want to add.
Small tweaks should be easy, and big new features that were not part of the design should not be too scary due to the existing code.
11
58
4
331
DALL-E empowered a 4-year old to design these red chicken slippers, which now exist in reality. Best kind of Sim2Real result:
My 4 year old designed cute fuzzy red chicken slippers with #dalle a couple weeks ago. 🤩
I had them custom made and they just arrived. Soooo cute 🥰
She named them Henry and Wendy 🐓🐓
This is the future of creativity, young entrepreneurship, fashion design and ecommerce.
7
22
1
157