Scott Gray
Scott Gray is a GPU engineer and research co-author who has worked at Nervana Systems and OpenAI. Before OpenAI's August 2016 team update discussed his work, he had optimized deep networks on GPUs at Nervana Systems. In 2017 he told The Register that OpenAI's block sparse GPU kernels could extend to other architectures. He co-authored the 2019 Sparse Transformers paper, the 2020 Language Models are Few-Shot Learners paper, and the 2021 Zero-Shot Text-to-Image Generation paper. In August 2026 reports said he had left OpenAI, and he now works independently on neuroscience-inspired AI methods.
- Works
- 12
Publication numbers from OpenAlex, as of 15 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.
Timeline
-
2016
GPU work at Nervana Systems
OpenAI's team update of August 2016 discusses his work on deep-network performance on GPUs at Nervana Systems.
-
2017
Block sparse GPU kernels
He tells The Register in 2017 that OpenAI's block sparse GPU kernels could extend to other architectures.
-
2019
Sparse Transformers
The 2019 paper Generating Long Sequences with Sparse Transformers introduces sparse factorizations of the attention matrix, and he co-authors it.
-
2020
Language models are few-shot learners
Language Models are Few-Shot Learners, which trains GPT-3, appears in the NeurIPS 2020 proceedings with him among its authors.
-
2021
Zero-shot text-to-image generation
A 2021 paper on generating images from text, Zero-Shot Text-to-Image Generation, lists him among its authors.
-
August 2026
Leaving OpenAI
Reports in August 2026 say he leaves OpenAI to explore neuroscience-inspired AI methods independently.
-
As of October 2026
Independent research after OpenAI
In 2026 he marks his OpenAI experience as former, and he works independently on neuroscience-inspired AI.
Sources and credits
Photo: We have no published photo of Scott Gray yet. Send us a free-licensed photo.
- OpenAI Loses Its "God of CUDA Kernels" - Exclusive Breakthrough on the High-Stakes Exit of Legendary Core Developer
- OpenAI uses cunning code to speed up GPU machine learning
- [1904.10509] Generating Long Sequences with Sparse Transformers
- Language Models are Few-Shot Learners
- [2102.12092] Zero-Shot Text-to-Image Generation