President & Co-Founder @OpenAI
Joined July 2010
- Tweets 4,815
- Following 65
- Followers 498,999
- Likes 1,386
DALL·E Outpainting — extend an existing image, to arbitrary size: openai.com/blog/dall-e-intro…
26
181
16
1,550
An overview of our approach to AGI alignment.
Highest-order bit is that we take an empirical approach — we deploy increasingly-powerful systems and get real-world feedback rather than try to solve everything a priori. Aim is to keep contact with reality.
openai.com/blog/our-approach…
6
14
2
99
DALL-E fashion show:
AI fashion show using dall-e to generate hundreds of outfits of the day after tomorrow. An interesting way to brainstorm costume and fashion design ideas.
Collab with @shyamagolden
@OpenAI #dalle2 #dalle #aiart
12
49
7
450
An application of Codex which "literally changed what we believe is possible" for a team faced with a rewrite from Perl to Python:
netapp.com/blog/open-ai-code…
10
22
8
166
Just launched a content moderation endpoint — uses our models to assess whether some text is sexual, hateful, violent, or promotes self-harm. Part of a strategy of scalable oversight of language models:
Introducing a free new tool that gives OpenAI API developers stronger content moderation abilities openai.com/blog/new-and-impr…
8
8
1
86
Applications for OpenAI Residency — a program for people without a machine learning background to enter the field — are open from now until Sept 30.
Looking for software engineers and researchers from other fields to apply their skills to deep learning:
The OpenAI Residency applications are open! We’re looking for engineers and researchers who are interested in applying their skills to AI and machine learning. Apply through Sept. 30.
Software Engineering: boards.greenhouse.io/openai/…
Research: boards.greenhouse.io/openai/…
13
30
3
173
Greg Brockman retweeted
Really proud to highlight work by @mobav0 and Heewoo showing that LMs can learn to infill text *for free*. Usually, when training on a mixture of tasks for a fixed token budget, you expect worse results compared to training fully on one of the tasks. 1/
Efficient Training of Language Models to Fill in the Middle
Shows that autoregressive LMs can learn to infill text after we apply a straightforward transformation to the dataset, which simply moves a span of text from the middle of a document to its end.
arxiv.org/abs/2207.14255
5
15
106
There is something really beautiful about code which has stood the test of time in production — you can't be sure it's bug free, but you can be certain that its usefulness outshines any flaws.
12
19
1
191
Coding conventions are often far less important than coding standards — that is, most approaches for organizing code can work, but the key is getting everyone to do it the same way throughout your codebase. Explains how there can be so many mutually-conflicting coding dogmas.
11
20
1
147
We are scaling DALL·E to 1 million people, and have announced beta pricing (in addition to free credits that refill automatically each month):
DALL·E 2 is officially now in beta. We’ll be inviting 1 million people from our waitlist over the coming weeks. openai.com/blog/dall-e-now-a…
21
19
2
221
Reducing bias in DALL·E, for prompts that do not specify race or gender:
DALL·E now generates images of people that more accurately reflect the diversity of the world’s population. openai.com/blog/reducing-bia…
7
6
52
Neural networks have a very convenient API from a systems perspective — constant time regardless of inputs, largely functional with explicit state, and decomposable into well-defined components which can be sharded across machines. Catnip for the systems engineer.
14
21
203
The challenge with debugging is purging your mind of what you think you know & being willing to face cold, hard reality. Requires great effort, but the reward (in addition to a working program) is a more accurate understanding of reality.
14
35
6
287