PeopleofSuperintelligence

Researcher

Erich Elsen

Also known as E ELSEN

Erich Elsen is an AI researcher whose work spans GPU-efficient neural networks, large language models and enterprise document reasoning. He studied at the University of Illinois and Stanford, then coauthored work on persistent and sparse GPU kernels and the compute-optimal Chinchilla study. In 2022, he co-founded Adept; he later coauthored the Persimmon-8B and Fuyu-8B models and joined Amazon’s Artificial General Intelligence unit in 2024. He then coauthored Databricks’ OfficeQA benchmark and Document Intelligence work; as of October 2026, he is a Principal Research Scientist at Databricks.

Works
49
Badges

Publication numbers from OpenAlex, as of 22 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

  1. 2004

    Illinois class of 2004

    He is a member of the University of Illinois Class of 2004 and spends countless hours practicing piano in Smith Memorial Hall. He goes on to study mechanical engineering at Stanford.

  2. 2016–2020

    Persistent RNNs and Fast Sparse ConvNets

    In 2016, he coauthors Persistent RNNs, a paper that maps deep recurrent neural networks onto GPUs and uses persistent kernels to raise computational throughput. In 2020, he coauthors Fast Sparse ConvNets, which introduces efficient sparse kernels for deep learning.

  3. 2022

    Chinchilla earns a NeurIPS award

    In 2022, he coauthors Training Compute-Optimal Large Language Models, which finds that model size and training tokens should scale equally for a fixed compute budget. NeurIPS names the compute-optimal training work an Outstanding Paper under the title An empirical analysis of compute-optimal large language model training.

  4. 2022

    Co-founding Adept

    Before he co-founds Adept in 2022, he works as a researcher at DeepMind for three years and as a software engineer at Google for two years.

  5. 2023

    Persimmon and Fuyu models

    In 2023, he coauthors Adept’s Persimmon-8B, a decoder model with approximately 8 billion parameters, and Fuyu-8B, a multimodal model that handles text and images.

  6. 2024

    Joining Amazon’s AGI group

    In 2024, he joins Amazon’s Artificial General Intelligence business unit as a Distinguished Scientist after co-founding Adept.

  7. December 2025–April 2026

    OfficeQA and document intelligence

    In December 2025, he coauthors OfficeQA, a benchmark for grounded reasoning over enterprise documents and tables. In April 2026, he coauthors Databricks’ Document Intelligence announcement for agentic workflows over business documents.

  8. As of October 2026

    Principal research scientist at Databricks

    As of October 2026, he is a Principal Research Scientist at Databricks.

Career

  1. NowDatabricksPrincipal Research Scientist
  2. From 2024AmazonAdept co-founder
  3. From 2022AdeptCo-founder
  4. Google DeepMindResearcher
  5. GoogleSoftware engineer

Education

  1. 2004University of Illinois Urbana-Champaign
  2. Stanford University

From the biography.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Erich Elsen. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  2. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Erich Elsen. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.
  3. Among employers with 30+ full-time cases, OpenAI leads primary technical role H-1B pay at $310,000, and Amazon, Google and Microsoft file 57% of all casesIncludes Databricks. Across 440,669 certified US H-1B filings from 44 employers (FY2020–FY2026), OpenAI leads primary technical role base pay at $310,000 and xAI leads overall base pay at $290,000.
  4. 19% of people shaping AI who founded a dated organization have founded two or more, waiting a median of 7 years before a secondIncludes Databricks. Across 1,136 people shaping AI who founded or co-founded a dated company, standalone lab or nonprofit, 221 (19%) founded at least two organizations, accounting for 36% of all 1,421 founding events.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

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

  1. Erich Elsen Profile | University of Illinois 150 Years uofi150.news-gazette.com. Checked 22 October 2026.
  2. Persistent RNNs: Stashing Recurrent Weights On-Chip proceedings.mlr.press. Checked 22 October 2026.
  3. Fast Sparse ConvNets openaccess.thecvf.com. Checked 22 October 2026.
  4. Training Compute-Optimal Large Language Models arxiv.org. Checked 22 October 2026.
  5. NeurIPS 2022 Awards neurips.cc. Checked 22 October 2026.
  6. 10 Huge AWS Exits And Hires From Google, OpenAI And Nvidia crn.com. Checked 22 October 2026.
  7. Persimmon | Transformers huggingface.co. Checked 22 October 2026.
  8. Fuyu | Transformers huggingface.co. Checked 22 October 2026.
  9. Introducing OfficeQA: A benchmark for end-to-end grounded reasoning | Databricks Blog databricks.com. Checked 22 October 2026.
  10. Why your agents can't read enterprise documents — and how to fix it | Databricks Blog databricks.com. Checked 22 October 2026.
  11. Articles by Erich Elsen - Databricks Blog databricks.com. Checked 22 October 2026.