President & Co-Founder @OpenAI
Joined July 2010
- Tweets 4,815
- Following 65
- Followers 495,334
- Likes 1,386
A fun proof-of-concept exploration of using GPT-3 in your daily life. Still a long way to go, but one day applications like this should be very useful:
Watch as we explore @OpenAI’s GPT-3 decision making ability. More playing around and experimentation with this intelligent machine to come soon! (hint: we have several series planned 👀)
11
20
1
86
GPT-3 for semantic search:
In 2020, Algolia partnered with @OpenAI to connect #GPT3 with its existing search technology.
The result: Algolia Answers, a next-generation product enabling intelligent, semantics-driven single-search endpoint for search queries.
@dcoates explains ▶️ bit.ly/3O1TvsB
7
6
46
DALL·E 2 + universe =
4
4
62
Using GPT-3 to explain how a given piece of code works:
Wrote up some notes on how I've been using GPT-3 as a tool for explaining how code works simonwillison.net/2022/Jul/9…
6
12
83
Never write a regex by hand again. An application I've wanted myself since about the time I wrote my first regex:
built a quick app that uses gpt-3 to convert from English to RegEx so you don't have to waste time on stackoverflow: autoregex.xyz
22
188
10
1,949
Give a machine a program, it will solve the problem of today. Teach a machine to program, it will solve the problems of a lifetime.
64
98
38
743
Nice piece in today's New Yorker on DALL·E. Found their image citation format really thoughtful & interesting:
'“a real hippopotamus sitting on a sofa smoking a cigarette, 70mm Nikon,” as prompted by Roz Chast. Art work by DALL-E / Courtesy OpenAI'
newyorker.com/magazine/2022/…
6
15
2
81
Even at OpenAI scale, systems optimizations need to be weighed against usability cost. A few percent improvement does not matter if the system is too hard to co-evolve with the ML. Trick is to do inordinate design thinking to unify hackability with the long-tail of optimizations.
6
19
1
139
We designed & implemented a variety of guardrails to align DALL·E 2's behavior to our content policies. A great post by @unixpickle describing how our pre-training mitigations work. Applicable to future powerful generative models as well:
openai.com/blog/dall-e-2-pre…
13
30
4
181
Applies to discussion about AGI. Far too easy to get caught up in quibbles over words, or to reach seemingly irrefutable conclusions based on implicit assumptions.
To make conceptual progress on how AGI might look, I try reason from the delta of what we know now vs in 2012.
When a debate descends to talking about the meanings of words, it's irrecoverable. You will learn nothing from continuing. I'm not sure why this is true, but empirically I've found to be true 100% of the time.
1
9
59
Machine learning is a great field for people who derive unreasonable pleasure from watching progress meters — there's a lot of joy to be found in watching each new model's learning curve go down and to the right.
18
60
10
667
Went to sleep at 3:30a last night trying to track down a nasty CUDA illegal memory access bug, woke up early to keep digging, & just now pushed a fix. Hard to beat the endorphin rush of doing battle with a tricky bug and coming out on top.
29
22
7
949