Safely aligning powerful AI systems is one of the most important unsolved problems for our mission. Techniques like learning from human feedback are helping us get closer, and we are actively researching new techniques to help us fill the gaps.
Focus areas
We build our generative models using a technology called deep learning, which leverages large amounts of data to train an AI system to perform a task.
Text
Our text models are advanced language processing tools that can generate, classify, and summarize text with high levels of coherence and accuracy.
Aligning language models to follow instructions
We’ve trained language models that are much better at following user intentions than GPT-3.Summarizing books with human feedback
We've trained a model to summarize entire books with human feedback.Language models are few-shot learners
We trained GPT-3, an autoregressive language model with 175 billion parameters.
Image
Our research on generative modeling for images has led to representation models like CLIP, which makes a map between text and images that an AI can read, and DALL-E, a tool for creating vivid images from text descriptions.
Hierarchical text-conditional image generation with CLIP latents
We show that explicitly generating image representations improves image diversity with minimal loss in photorealism and caption similarity.DALL·E: Creating images from text
We’ve trained a neural network called DALL·E that creates images from text captions for a wide range of concepts expressible in natural language.CLIP: Connecting text and images
We’re introducing a neural network called CLIP which efficiently learns visual concepts from natural language supervision.
Audio
Our research on applying AI to audio processing and audio generation has led to developments in automatic speech recognition and original musical compositions.
Introducing Whisper
We’ve trained and are open-sourcing a neural net that approaches human level robustness and accuracy on English speech recognition.Jukebox
We’re introducing Jukebox, a neural net that generates music as raw audio in a variety of genres and artist styles.MuseNet
We’ve created MuseNet, a deep neural network that can generate 4-minute musical compositions with 10 different instruments.

Past highlights
Our current AI research builds upon a wealth of previous projects and advances.
Featured roles
We are constantly seeking talented individuals to join our team. Explore featured roles or view all open roles.