Prafulla Dhariwal

Early work

Prafulla Dhariwal works at OpenAI on generative models and unsupervised learning. Earlier papers Dhariwal co-authored include "Proximal Policy Optimization Algorithms" (2017) with John Schulman and others, and Glow (2018), a flow-based generative model with invertible 1x1 convolutions, written as co-first author with Diederik P. Kingma.¹¹²⁹

Dhariwal was first author of Jukebox (2020), a generative model that produces music with singing as raw audio, and appears among the authors of "Language Models are Few-Shot Learners" (2020), the GPT-3 paper.⁸¹

Diffusion models and image generation

With Alex Nichol, Dhariwal co-first-authored "Improved Denoising Diffusion Probabilistic Models" (ICML 2021) and "Diffusion Models Beat GANs on Image Synthesis" (NeurIPS 2021), and OpenAI released MIT-licensed code and checkpoints for both. Dhariwal was also a co-first author of GLIDE (2021), a text-guided diffusion model for image generation and editing.²³⁴⁵⁶⁷

Dhariwal was a co-first author of the DALL·E 2 paper, "Hierarchical Text-Conditional Image Generation with CLIP Latents" (2022), co-authored Point·E (2022) for generating 3D point clouds from text prompts, and is listed among the authors of DALL·E 3. With Yang Song, Dhariwal co-authored "Consistency Models" (2023) and "Improved Techniques for Training Consistency Models" (2023), which generate samples in one or a few steps.^13141101511

Work and contributions

• Guided diffusion (Diffusion Models Beat GANs on Image Synthesis) Open-source code

  Co-first author with Alex Nichol of the 2021 paper. The openai/guided-diffusion repository is the paper's codebase, builds on openai/improved-diffusion from the pair's Improved DDPM paper, releases checkpoints for the main models, and is licensed under the MIT License.³²⁵⁶

• Jukebox Published research

  First author of "Jukebox: A Generative Model for Music" (2020), with Heewoo Jun, Christine Payne, Jong Wook Kim, Alec Radford, and Ilya Sutskever.⁸¹

• Consistency models Open-source code

  Co-author with Yang Song, Mark Chen, and Ilya Sutskever of "Consistency Models" (2023) and, with Yang Song, of "Improved Techniques for Training Consistency Models" (2023). The openai repository releases code and checkpoints under the MIT License.¹⁰¹⁵¹¹

Timeline

1. Jul 20, 2017Co-authored "Proximal Policy Optimization Algorithms" with John Schulman and others.¹²
2. Jul 9, 2018Co-first author with Diederik P. Kingma of Glow, a flow-based generative model.⁹
3. Apr 30, 2020First author of Jukebox, a generative model for music.⁸
4. Feb 18, 2021Co-first author with Alex Nichol of "Improved Denoising Diffusion Probabilistic Models".²
5. May 11, 2021Co-first author with Alex Nichol of "Diffusion Models Beat GANs on Image Synthesis".³
6. Dec 20, 2021Co-first author of GLIDE, a text-guided diffusion model for photorealistic image generation and editing.⁷
7. Apr 13, 2022Co-first author of the DALL·E 2 paper on text-conditional image generation with CLIP latents.¹³
8. Dec 16, 2022Co-authored Point·E, a system for generating 3D point clouds from complex prompts.¹⁴
9. Mar 2, 2023Co-authored "Consistency Models" with Yang Song, Mark Chen and Ilya Sutskever.¹⁰
10. Oct 22, 2023Co-authored "Improved Techniques for Training Consistency Models" with Yang Song.¹⁵

Where they work

• OpenAI 63 people in the wiki

Related people

• Yang SongAlso at OpenAI
• Pieter AbbeelAlso at OpenAI
• Aditya RameshAlso at OpenAI
• Diederik P. KingmaAlso at OpenAI
• Jong Wook KimAlso at OpenAI
• Jade CopetImage, video, and 3D generation · Speech and audio
• Heiga ZenSpeech and audio · Image, video, and 3D generation
• Jonathan ShenSpeech and audio · Image, video, and 3D generation
• Neil ZeghidourSpeech and audio · Image, video, and 3D generation
• Sanyuan ChenSpeech and audio · Image, video, and 3D generation