Aligning AI systems with human intent

OpenAI’s mission is to ensure that artificial intelligence benefits all of humanity. An important part of this effort is training AI systems to do what humans want.

Our approach to alignment research August 24 update Read now
play Watch video
OpenAI researchers and staff, in order of appearance: Mira Murati, Jan Leike, Ryan Lowe, Jonathan Ward, William Saunders, Tyna Eloundou Nekoul, Sandhini Agarwal, Jeff Wu, Bianca Martin & Carroll Wainwright.

OpenAI’s Alignment research focuses on training AI systems to be helpful, truthful, and safe. Our team is exploring and developing methods to learn from human feedback.

Our long-term goal is to achieve scalable solutions that will align far more capable AI systems of the future — a critical part of our mission.

Learn More
AI-Written Critiques Help Humans Notice Flaws
Showing model-generated critical comments to humans helps them find flaws in summaries. Read more
Aligning Language Models to Follow Instructions
We trained language models that are much better than GPT-3 at following user intentions while also making them more truthful and less toxic. Read more
Learning to Summarize with Human Feedback
We applied reinforcement learning from human feedback to train language models that are better at summarization. Read more
More Alignment Research
Summarizing Books with Human Feedback
Learning Complex Goals with Iterated Amplification
AI Safety via Debate
Learning from Human Preferences
Other OpenAI Safety Projects
WebGPT: Improving the Factual Accuracy of Language Models through Web Browsing
Improving Language Model Behavior by Training on a Curated Dataset
Multimodal Neurons in Artificial Neural Networks
OpenAI Microscope
Safety Gym