Jaime Sevilla

Career

Sevilla completed a double degree in mathematics and computer engineering at Universidad Complutense de Madrid in 2019 and worked as a deep learning specialist on a high-energy astrophysics research programme. From 2020 Sevilla pursued a PhD at the University of Aberdeen as an NL4XAI early-stage researcher on explainability in Bayesian networks, paused in 2022.⁴

Sevilla co-founded Epoch AI in 2022 and leads it as Chief Executive Officer. From 2021 to 2024 Sevilla was also interim director and co-founder of the Observatorio de Riesgos Catastróficos Globales, a science-policy organization, and from 2019 to 2025 a research affiliate at the Centre for the Study of Existential Risk at the University of Cambridge.⁴¹

Research

Sevilla's work at Epoch AI studies what drives progress in machine learning. "Compute Trends Across Three Eras of Machine Learning" (2022), with Sevilla as first author, traced the growth of training compute, and Sevilla co-authored "Will we run out of data?" on limits of scaling based on human-generated text.⁵⁷³

In 2024 Sevilla and Edu Roldán published an update on training-compute growth based on Epoch AI's expanded AI model database, and Sevilla led the report "Can AI Scaling Continue Through 2030?", which examined constraints including power, chip production and data. Sevilla also co-authored "Please Report Your Compute" in Communications of the ACM (2023).⁸⁶³

Work and contributions

• Epoch AI Data on AI Models Open data

  Co-authored, with Edu Roldán, a 2024 Epoch AI report that used the organization's expanded AI model database to analyse growth in the compute used to train notable models.⁸

• Compute Trends Across Three Eras of Machine Learning Published research

  First author of a 2022 paper analysing the history of training compute for machine learning systems across three eras.⁵³

• Can AI Scaling Continue Through 2030? Published research

  Lead author of an Epoch AI report examining constraints on AI training scale up to 2030, such as power availability and chip manufacturing.⁶

Timeline

1. 2020Began a PhD at the University of Aberdeen as an NL4XAI early-stage researcher.⁴
2. 2022Co-founded Epoch AI as director.⁴
3. Feb 2022Posted "Compute Trends Across Three Eras of Machine Learning" as first author.⁵
4. Oct 2022Co-authored "Will we run out of data?" on limits of LLM scaling based on human-generated data.⁷
5. May 28, 2024Published an Epoch AI analysis of training-compute growth based on the expanded AI model database.⁸
6. Aug 20, 2024Published "Can AI Scaling Continue Through 2030?" as lead author.⁶

Other documented roles

Roles the sources document that are not confirmed as current.

• Interim Director and Co-Founder (2021–2024), Observatorio de Riesgos Catastróficos Globales⁴

Where they work

• Epoch AI 2 people in the wiki

Related people

• Marius HobbhahnAlso at Epoch AI
• Pradeep DasigiEvaluation and benchmarks · Datasets and data
• Timnit GebruDatasets and data · Evaluation and benchmarks
• Kyle LoDatasets and data · Evaluation and benchmarks
• Anton TroynikovDatasets and data · Evaluation and benchmarks
• Adina WilliamsEvaluation and benchmarks · Datasets and data
• Ruchir PuriDatasets and data · Evaluation and benchmarks
• Jeff HuberDatasets and data · Evaluation and benchmarks
• Jesse DodgeEvaluation and benchmarks · Datasets and data
• Sasha LuccioniEvaluation and benchmarks · Datasets and data