Large Language Models

It’s been a year of dramatic change and growth at OpenAI.

With diverse environments, we can analyze, diagnose and edit deep reinforcement learning models using attribution.

OpenAI has agreed to license GPT-3 to Microsoft for their own products and services.

A collection of articles and comments with the goal of understanding how to design robust and general purpose self-organizing systems.

Our third class of OpenAI Scholars presented their final projects at virtual Demo Day, showcasing their research results from over the past five months.

We’re excited to announce that OpenAI is co-organizing two NeurIPS 2020 competitions with AIcrowd, Carnegie Mellon University, and DeepMind, using Procgen Bench

We’re releasing an API for accessing new AI models developed by OpenAI.

[Updated on 2020-06-17: Add “exploration via disagreement” in the “Forward Dynamics” section . Exploitation versus ex

We’re releasing an analysis showing that since 2012 the amount of compute needed to train a neural net to the same performance on ImageNet classification has be

We’re introducing Jukebox, a neural net that generates music, including rudimentary singing, as raw audio in a variety of genres and artist styles. We’re releas

We’ve contributed to a multi-stakeholder report by 58 co-authors at 30 organizations, including the Centre for the Future of Intelligence, Mila, Schwartz Reisma

We are standardizing OpenAI’s deep learning framework on PyTorch.

We’re releasing Procgen Benchmark, 16 simple-to-use procedurally-generated environments which provide a direct measure of how quickly a reinforcement learning a

We’re releasing Safety Gym, a suite of environments and tools for measuring progress towards reinforcement learning agents that respect safety constraints while

[Updated on 2020-01-09: add a new section on Contrastive Predictive Coding ]. [Updated on 2020-04-13: add a “Momentum Contra

As the final model release of GPT-2’s staged release, we’re releasing the largest version (1.5B parameters) of GPT-2 along with code and model weights to facili

We’ve trained a pair of neural networks to solve the Rubik’s Cube with a human-like robot hand. The neural networks are trained entirely in simulation, using th

We are now accepting applications for our third class of OpenAI Scholars.

A closer look at how Temporal Difference Learning merges paths of experience for greater statistical efficiency

We’ve fine-tuned the 774M parameter GPT-2 language model using human feedback for various tasks, successfully matching the preferences of the external human lab

Stochastic gradient descent is a universal choice for optimizing deep learning models. However, it is not the only option. With bl

We’ve developed a method to assess whether a neural network classifier can reliably defend against adversarial attacks not seen during training. Our method yiel

We’re releasing the 774 million parameter GPT-2 language model after the release of our small 124M model in February, staged release of our medium 355M model in

Section 3.2 of Ilyas et al. (2019) shows that training a model on only adversarial errors leads to non-trivial generalization on the original test set. We show

Microsoft is investing $1 billion in OpenAI to support us building artificial general intelligence (AGI) with widely distributed economic benefits. We’re partne

We’ve written a policy research paper identifying four strategies that can be used today to improve the likelihood of long-term industry cooperation on safety n

In my earlier post on meta-learning , the problem is mainly defined in the context of few-shot classificati

We hosted the first OpenAI Robotics Symposium on April 27, 2019.

Our second class of OpenAI Scholars has concluded, with all eight scholars producing an exciting final project showcased at Scholars Demo Day at OpenAI.

Our second class of OpenAI Fellows has wrapped up, with each Fellow going from a machine learning beginner to core OpenAI contributor in the course of a 6-month

In Robotics, one of the hardest problems is how to make your model transfer to the real world. Due to the sample inefficiency of deep RL algorithms and the cost

We’ve created MuseNet, a deep neural network that can generate 4-minute musical compositions with 10 different instruments, and can combine styles from country

We’ve developed the Sparse Transformer, a deep neural network which sets new records at predicting what comes next in a sequence—whether text, images, or sound.

OpenAI Five is the first AI to beat the world champions in an esports game, having won two back-to-back games versus the world champion Dota 2 team, OG, at Fina

We’ll be holding our final live event for OpenAI Five at 11:30am PT on April 13.

We’ve made progress towards stable and scalable training of energy-based models (EBMs) resulting in better sample quality and generalization ability than existi

Our class of eight scholars (out of 550 applicants) brings together collective expertise in literature, philosophy, cell biology, statistics, economics, quantum

We’ve created activation atlases (in collaboration with Google researchers), a new technique for visualizing what interactions between neurons can represent. As

We’re releasing a Neural MMO, a massively multiagent game environment for reinforcement learning agents. Our platform supports a large, variable number of agent

On February 2, we held our first Spinning Up Workshop as part of our new education initiative at OpenAI.

If we want to train AI to do what humans want, we need to study humans.

We’ve written a paper arguing that long-term AI safety research needs social scientists to ensure AI alignment algorithms succeed when actual humans are involve