PeopleofSuperintelligence

Researcher

Jack Rae

Jack Rae is an AI researcher whose work spans neural memory, large language models and model reasoning. In 2016 he coauthored Sparse Access Memory, and in 2017 he helped open-source DeepMind's Sonnet library. His DeepMind research included Gopher and Chinchilla; the latter received a 2022 NeurIPS Outstanding Paper Award. He coauthored OpenAI's GPT-4 technical report in 2023 and discussed thinking models, inference time and long context as a Google DeepMind Principal Scientist in February 2025. He is an AI researcher at Meta Superintelligence Labs, working on frontier-model training and reasoning.

Works
31
Citations
3,891
h-index
22

Publication numbers from OpenAlex, as of 15 October 2026.

Timeline

  1. 2016–2017

    Sparse Access Memory and Sonnet

    Rae and his coauthors introduce Sparse Access Memory in 2016, an external memory system with sparse reads and writes. In 2017, he coauthors DeepMind's announcement of Sonnet, an open-source TensorFlow library for neural-network modules.

  2. 2021

    Scaling Gopher

    At DeepMind, he coauthors a study of Gopher, a 280-billion-parameter transformer language model. The team compares models at different scales and reports gains in reading comprehension and fact-checking.

  3. 2022

    Compute-optimal Chinchilla

    In 2022, he coauthors the compute-optimal training paper that receives a NeurIPS Outstanding Paper Award. Chinchilla uses 70 billion parameters and four times more data than Gopher for the same compute budget.

  4. March 2023

    GPT-4 technical report

    He coauthors OpenAI's GPT-4 technical report in March 2023. The report describes a multimodal model that accepts image and text inputs and performs at a human level on professional and academic benchmarks.

  5. February 2025

    Thinking models at Google

    He is a Principal Scientist at Google DeepMind in February 2025. On Google's AI podcast, he discusses thinking models, increased thinking time and long context.

  6. Since 2025

    AI researcher at Meta

    He joins Meta Superintelligence Labs in June 2025, after leading Gemini pre-training and Gemini 2.5 reasoning work at Google DeepMind. In 2026, he is an AI researcher at Meta, working on frontier-model training and reasoning.

Sources and credits

Photo: We have no published photo of Jack Rae yet. Send us a free-licensed photo.

  1. Scaling Memory-Augmented Neural Networks with Sparse Reads and Writes papers.neurips.cc. Checked 14 October 2026.
  2. Open sourcing Sonnet - a new library for constructing neural networks — Google DeepMind deepmind.google. Checked 14 October 2026.
  3. Language modelling at scale: Gopher, ethical considerations, and retrieval — Google DeepMind deepmind.google. Checked 14 October 2026.
  4. NeurIPS 2022 Awards neurips.cc. Checked 14 October 2026.
  5. GPT-4 Technical Report arxiv.org. Checked 14 October 2026.
  6. Google AI: Release Notes podcast about thinking models blog.google. Checked 14 October 2026.
  7. Mark Zuckerberg creating Meta Superintelligence Labs. Read the memo cnbc.com. Checked 14 October 2026.
  8. Here Is Everyone Mark Zuckerberg Has Hired So Far for Meta’s ‘Superintelligence’ Team | WIRED wired.com. Checked 14 October 2026.
  9. Jack Rae: Bio, Work & Ideas | AI Engineer ai.engineer. Checked 14 October 2026.
  10. Meta Superintelligence Labs: What We Know So Far | Built In builtin.com. Checked 14 October 2026.