Blog

Research

2022
Introducing Whisper
2022
Our approach to alignment research
2022
DALL·E 2 Pre-Training Mitigations
2022
Learning to Play Minecraft with Video PreTraining (VPT)
2022
AI-Written Critiques Help Humans Notice Flaws
2022
Techniques for Training Large Neural Networks
2022
Measuring Goodhart’s Law
2022
DALL·E 2
2022
Lessons Learned on Language Model Safety and Misuse
2022
Solving (Some) Formal Math Olympiad Problems
2022
Aligning Language Models to Follow Instructions
2022
Introducing Text and Code Embeddings in the OpenAI API
2021
WebGPT: Improving the Factual Accuracy of Language Models through Web Browsing
2021
Solving Math Word Problems
2021
Summarizing Books with Human Feedback
2021
Introducing Triton: Open-Source GPU Programming for Neural Networks
2021
Improving Language Model Behavior by Training on a Curated Dataset
2021
Multimodal Neurons in Artificial Neural Networks
2021
Scaling Kubernetes to 7,500 Nodes
2021
DALL·E: Creating Images from Text
2021
CLIP: Connecting Text and Images
2020
Learning to Summarize with Human Feedback
2020
Image GPT
2020
AI and Efficiency
2020
Jukebox
2020
Improving Verifiability in AI Development
2020
OpenAI Microscope
2020
OpenAI Standardizes on PyTorch
2019
Deep Double Descent
2019
Procgen Benchmark
2019
Safety Gym
2019
GPT-2: 1.5B Release
2019
Solving Rubik’s Cube with a Robot Hand
2019
Fine-Tuning GPT-2 from Human Preferences
2019
Emergent Tool Use from Multi-Agent Interaction
2019
Testing Robustness Against Unforeseen Adversaries
2019
GPT-2: 6-Month Follow-Up
2019
Why Responsible AI Development Needs Cooperation on Safety
2019
MuseNet
2019
Generative Modeling with Sparse Transformers
2019
OpenAI Five Defeats Dota 2 World Champions
2019
OpenAI Five Finals
2019
Implicit Generation and Generalization Methods for Energy-Based Models
2019
Introducing Activation Atlases
2019
Neural MMO: A Massively Multiagent Game Environment
2019
AI Safety Needs Social Scientists
2019
Better Language Models and Their Implications
2018
How AI Training Scales
2018
Quantifying Generalization in Reinforcement Learning
2018
Spinning Up in Deep RL
2018
Learning Concepts with Energy Functions
2018
Reinforcement Learning with Prediction-Based Rewards
2018
Learning Complex Goals with Iterated Amplification
2018
The International 2018: Results
2018
OpenAI Five Benchmark: Results
2018
Learning Dexterity
2018
Glow: Better Reversible Generative Models
2018
Learning Montezuma's Revenge from a Single Demonstration
2018
OpenAI Five
2018
Retro Contest: Results
2018
Improving Language Understanding with Unsupervised Learning
2018
Gym Retro
2018
AI and Compute
2018
AI Safety via Debate
2018
Evolved Policy Gradients
2018
Retro Contest
2018
Reptile: A Scalable Meta-Learning Algorithm
2018
Ingredients for Robotics Research
2018
Preparing for Malicious Uses of AI
2018
Interpretable Machine Learning through Teaching
2018
Discovering Types for Entity Disambiguation
2018
Requests for Research 2.0
2018
Scaling Kubernetes to 2,500 Nodes
2017
Block-Sparse GPU Kernels
2017
Learning a Hierarchy
2017
Generalizing from Simulation
2017
Meta-Learning for Wrestling
2017
Competitive Self-Play
2017
Nonlinear Computation in Deep Linear Networks
2017
Learning to Model Other Minds
2017
OpenAI Baselines: ACKTR & A2C
2017
More on Dota 2
2017
Dota 2
2017
Gathering Human Feedback
2017
Better Exploration with Parameter Noise
2017
Proximal Policy Optimization
2017
Robust Adversarial Inputs
2017
Faster Physics in Python
2017
Learning from Human Preferences
2017
Learning to Cooperate, Compete, and Communicate
2017
OpenAI Baselines: DQN
2017
Robots that Learn
2017
Roboschool
2017
Unsupervised Sentiment Neuron
2017
Spam Detection in the Physical World
2017
Evolution Strategies as a Scalable Alternative to Reinforcement Learning
2017
Distill
2017
Learning to Communicate
2017
Attacking Machine Learning with Adversarial Examples
2016
Faulty Reward Functions in the Wild
2016
Universe
2016
Infrastructure for Deep Learning
2016
Special Projects
2016
Concrete AI Safety Problems
2016
Generative Models
2016
OpenAI Gym Beta