President & Co-Founder @OpenAI
Joined July 2010
- Tweets 4,815
- Following 65
- Followers 495,334
- Likes 1,386
Our robotic system solves the cube 60% of the time under normal conditions — but only 20% with an adversarially-scrambled cube.
The big result is that it's possible at all. Like with OpenAI Five, reliability keeps getting better the more we train
This tweet is unavailable
5
4
2
48
Yes. The reason this result is exciting is that the technology is general-purpose — it can be applied to any simulatable task.
1
2
The robot didn't get to train *at all* with tied fingers — it had to adapt on the fly.
(Also, humans have a billion plus years of evolutionary practice to solve the cube with untied fingers; the robot only gets about 10,000 years of untied practice.)
I could learn it in a year with all my fingers. Then it would probably only take a couple days to adjust my mental model to solve it with two fingers tied.
How many billion years of training did the Deep RL agent need again?
7
14
6
116
No special training here — the robot’s never seen anything like this and needs to adapt on the fly.
2
1
6
How well could *you* solve the the Rubik’s cube with two fingers tied together? (Especially without getting to practice first!)
7
17
6
171
0
Our robotic system can adapt to new situations it definitely didn't see this during training:
24
153
61
1,037
0
Announcing major progress towards general-purpose robots.
This robot is trained using the same reinforcement learning code as OpenAI Five, plus a new technique for transferring knowledge from simulation to reality.
Result is unprecedented dexterity:
We've trained an AI system to solve the Rubik's Cube with a human-like robot hand.
This is an unprecedented level of dexterity for a robot, and is hard even for humans to do.
The system trains in an imperfect simulation and quickly adapts to reality: openai.com/blog/solving-rubi…
9
101
10
524
Excited to announce that we're opening applications for our third class of OpenAI Scholars, a program providing mentorship and stipend to support people from underrepresented groups entering AI. We think diversity is core to AI being positive for everyone:
Now accepting applications for our 3rd class of OpenAI Scholars: a 4 month full-time program for individuals from underrepresented groups to study deep learning and produce an open-source project. Mentors include @mcleavey, @karlcobbe, @AlecRad: openai.com/blog/openai-schol…
2
6
45
A reporter's reaction GPT-2: "And because GPT-2 was an inspired mimic, expertly capturing The New Yorker’s cadences and narrative rhythms, it sounded like a familiar, trusted voice that I was inclined to believe. In fact, it sounded sort of like my voice."
Could a robot replace a New Yorker writer? We fed The New Yorker’s archive to an artificial-intelligence writer, which predicts text based on preceding language. Then we asked it to write for us. nyer.cm/5H11CgF
2
32
3
95
We tend to act as a seed crystal for new advances. There are now many popular, actively maintained GPT-2 code repositories (such as your very own github.com/minimaxir/gpt-2-s…) or interactive websites (talktotransformer.com). We focus on making the next breakthrough.
2
6
Pretty catchy song composed by MuseNet, with lyrics by GPT-2, performed by a human: youtube.com/watch?v=fN-bQddb…
2
10
1
66
That being said, this GPT-2 written argument certainly wasn't boring: nitter.vloup.ch/gdb/status/10960…!
An OpenAI employee printed out this AI-written sample and posted it by the recycling bin: blog.openai.com/better-langu…
1
19
.@karoly_zsolnai tested GPT-2: "It recognized that we are talking about fluid simulations, which is already remarkable, but it went much further. The completion is not bad at all and is not only coherent, on topic, but has quite a bit of truth to it."
2
3
35
A GPT-2 written essay was submitted to the Economist's youth essay contest.
One judge, who did not know the essay was written by an AI, gave this review: "It is strongly worded and backs up claims with evidence, but the idea is not incredibly original."
economist.com/open-future/20…
20
245
31
753
They’re all on Amazon! Here are two:
GPT-2 is now a published author:
amazon.com/gp/aw/sitb/B07X21…
amazon.com/gp/aw/sitb/B07X9F…
2
2
I’m starting a collection of books authored by GPT-2. Excited to add to the office bookshelf!
9
29
3
228