President & Co-Founder @OpenAI
Joined July 2010
- Tweets 4,815
- Following 65
- Followers 495,334
- Likes 1,386
Just released Jukebox, a neural network that generates music as raw audio. All of the sound you hear, including the singing, is generated:
Introducing Jukebox, a neural net that generates music, including rudimentary singing, as raw audio in a variety of genres and artist styles. We're releasing a tool for everyone to explore the generated samples, as well as the model and code: openai.com/blog/jukebox/
10
47
7
282
A cross-institutional report on making it easier to verify the properties of AI systems:
We've contributed to a multi-stakeholder report by 58 co-authors at 30 organizations that describes 10 mechanisms to improve the verifiability of claims made about AI systems. openai.com/blog/improving-ve…
1
9
25
We’ve released a tool for visualizing every layer and neuron in eight common vision networks. One more step towards opening the “black box” of the neural network.
Introducing OpenAI Microscope: a collection of visualizations of every layer and neuron in eight vision "model organisms" often studied in interpretability. This tool allows researchers to investigate down to individual units, and share those observations. openai.com/blog/microscope/
2
23
1
141
Neural nets are not black boxes. With study, we can interpret circuits they learn.
For example, the image below shows a car detector being composed of wheel, window, and car body detectors.
We show how to read a human-interpretable algorithm from a neural network's weights, by studying the circuits formed by the connections between individual neurons: distill.pub/2020/circuits/zo…
5
52
3
307
That'd be totally fine! Per our post: "Going forward we’ll primarily use PyTorch as our deep learning framework but sometimes use other ones when there’s a specific technical reason to do so".
1
25
OpenAI → PyTorch:
We're standardizing OpenAI's deep learning framework on PyTorch to increase our research productivity at scale on GPUs (and have just released a PyTorch version of Spinning Up in Deep RL): openai.com/blog/openai-pytor…
3
32
4
388
Watched "The Squire — AI Written Short Film".
It's a live-action adaption of AI Dungeon gameplay. Super fun to see!
youtube.com/watch?v=Kx-2Pyrh…
4
24
3
150
Welcome, @jeffclune!
I am extremely excited to announce (1) I've joined OpenAI to lead a large-scale effort into AI-generating Algorithms research, & (2) I'll be an Associate CS Professor at U. British Columbia in 2021, where I will continue to lead the OpenAI project. Both are dreams come true! 1/2
1
5
55
Working with Jakub Pachocki (@merettm) is one of my favorite parts of OpenAI. Highly recommend watching his talk!
Two updates on the deep learning lecture series starting Jan 6. If you are local to MIT & attending, we moved to a larger room: 54-100. Plus, new speaker added: Jakub Pachocki, research lead, @OpenAI Dota Team & Reasoning Team. Videos will be posted here: deeplearning.mit.edu
2
2
1
52
(Correction after talking to the artists: the lyrics were generated with GPT-2, but the melody was human-written, not MuseNet. Hopefully MuseNet next time :).)
3
11
A cool Christmas song, powered by GPT-2 and MuseNet:
christmas is dead. made a christmas song with AI. thanks @adamdanielking and @OpenAI soundcloud.com/bluethalo/chr…
5
8
29
I agree with this article that AI Dungeon 2 (the GPT-2 powered text-based game of infinite possibilities) is "one of the coolest video game experiments of 2019". You should definitely check it out: play.aidungeon.io/. Great work @nickwalton00!
11
87
6
430
Thanks!
I'm curious what reward function details you found in there that were missing from the description linked in the original OpenAI Five post: gist.github.com/dfarhi/66ec9…
1
1
3
Thank you! Means a lot coming from the best team in Dota 2 history :).
Wow! The paper looks insanely good!
Also nice try calling the new agent Rerun but we can't guarantee you the Bo3 would end the same way in case of a rematch 😏
#DreamOG
1
23
We just released our scientific analysis of OpenAI Five: nitter.vloup.ch/OpenAI/status/12…
We are already using findings from Five in other systems at OpenAI like Dactyl (openai.com/blog/solving-rubi…) or our multi-agent work (openai.com/blog/emergent-too…).
Hope that others find the results useful!
We're releasing "Dota 2 with Large Scale Deep Reinforcement Learning", a scientific paper analyzing our findings from our 3-year Dota project: openai.com/projects/five/
One highlight — we trained a new agent, Rerun, which has a 98% win rate vs the version that beat @OGEsports.
2
16
2
75