Draft:Epoch AI
Submission declined on 11 July 2026 by Zippybonzo (talk).
Where to get help
How to improve a draft
You can also browse Wikipedia:Featured articles and Wikipedia:Good articles to find examples of Wikipedia's best writing on topics similar to your proposed article. Improving your odds of a speedy review To improve your odds of a faster review, tag your draft with relevant WikiProject tags using the button below. This will let reviewers know a new draft has been submitted in their area of interest. For instance, if you wrote about a female astronomer, you would want to add the Biography, Astronomy, and Women scientists tags. Editor resources
|  |
| Founded | 2021 |
|---|---|
| Founders |
|
| Type | Nonprofit (501(c)(3)) |
Key people | Jaime Sevilla (co-founder and CEO) |
| Employees | 21 (2025) |
| Website | epoch |
Epoch AI is a quantitative research nonprofit that studies trends in artificial intelligence (AI) from an independent, evidence-based perspective, with the stated goal of helping people understand what is happening in AI "from a neutral perspective and grounded in the best possible evidence."[1] It operates as a distributed, remote-first organization with staff located globally.[1] Epoch AI was founded in 2021 as a volunteer group and formally established as a nonprofit in 2022, after its early research on trends in the computational resources ("compute") used to train machine learning models.[1][2]
It offers research services to companies, nonprofits, and government bodies, including commissioned research, model evaluations, and consultations.[3] It contributes data to Stanford University's AI Index, which draws on Epoch's datasets in its analysis of AI models, compute, and hardware, and it lists the United Kingdom's Department for Science, Innovation and Technology among its past and present partners.[1] The organization is an independent nonprofit funded by a combination of philanthropic donations and revenue from commissioned work.[4]
History
[edit]Epoch AI grew out of an informal, volunteer effort begun in 2021 and was established as a nonprofit in April 2022.[1][2] Its seven co-founders — Jaime Sevilla, Tamay Besiroglu, Lennart Heim, Pablo Villalobos, Edu Roldán, Marius Hobbhahn, and Anson Ho — had collective backgrounds spanning machine learning, statistics, economics, forecasting, physics, computer engineering, and software engineering.[1] The organization became an independent 501(c)(3) nonprofit in early 2025.[5] Sevilla, a Spanish researcher with a background in mathematics and computer science whose work focuses on technological forecasting and the trajectory of AI, serves as the organization's chief executive officer.[6][2] Founding member Lennart Heim, a researcher at the RAND Corporation, has remained affiliated with the organization.[2] In 2024, Sevilla was profiled by Time magazine for Epoch AI's work attempting to forecast the future development of AI.[2]
Research
[edit]Epoch AI's work spans empirical tracking of AI capabilities, analysis of the AI industry's economics and infrastructure, and independent model evaluations. The organization frames its research around three areas: the drivers of AI progress (such as training compute trends, inference scaling, and bottlenecks to continued scaling), the measurement of AI progress (through independent evaluations and custom benchmarks), and the impact of AI (including economic analysis and polling on adoption).[1][7]
Additional areas of study include semiconductor manufacturing and chip performance, data center development and capacity, energy consumption, the financial trajectories of leading AI companies, and the geopolitics of AI development.[7]
In a peer-reviewed 2024 study presented at the International Conference on Machine Learning, Epoch researchers projected that AI developers could exhaust the supply of publicly available human-generated text for training language models at some point between roughly 2026 and 2032. The study, which co-founder Tamay Besiroglu described as reflecting a "literal gold rush" for finite data, received wide media coverage.[8]
Benchmarks
[edit]Epoch AI develops independent benchmarks and evaluations of AI model capabilities. These include the Epoch Capabilities Index, a measure of model performance, and a public AI benchmarking hub.[7]
The organization is also known for FrontierMath, a benchmark consisting of extremely challenging mathematics problems, including some open research problems that remain unsolved by mathematicians. The problems were authored and peer-reviewed by expert mathematicians and range from advanced undergraduate material through research-level mathematics, with a separate collection of unsolved problems "that have resisted serious attempts by professional mathematicians."[9] The benchmark's creation was funded by OpenAI, which used FrontierMath in December 2024 to demonstrate the mathematical abilities of its o3 model. OpenAI has a verbal agreement with Epoch AI not to train its models on the benchmark's problems, and Epoch AI maintains a separate, unseen holdout set that allows it to independently verify results reported on FrontierMath.[10][11] Open Philanthropy (later Coefficient Giving) separately funded improvements to the benchmark, and a pilot version had earlier been funded by Nat Friedman.[12][11]
See also
[edit]External links
[edit]Category:Artificial intelligence associations
Category:Non-profit organizations
References
[edit]- 1 2 3 4 5 6 7 "About". Epoch AI. Retrieved 2026-07-08.
- 1 2 3 4 5 Henshall, Will (2024-06-05). "The Researcher Trying to Glimpse the Future of AI". Time. Retrieved 2026-07-08.
- ↑ Team, The Epoch AI. "Epoch AI 2025 impact report". Epoch AI. Retrieved 2026-09-14.
- ↑ AI Index Steering Committee (2026). Artificial Intelligence Index Report 2026 (PDF) (Report). Stanford Institute for Human-Centered Artificial Intelligence. Retrieved 2026-07-08.
- ↑ "Epoch AI 2025 impact report". Epoch AI. 2026-01-16. Retrieved 2026-07-08.
- ↑ "Jaime Sevilla". Epoch AI. Retrieved 2026-07-08.
- 1 2 3 "Epoch AI". Epoch AI. Retrieved 2026-07-08.
- ↑ O'Brien, Matt (2024-06-06). "AI 'gold rush' for chatbot training data could run out of human-written text as early as 2026". PBS NewsHour. Associated Press. Retrieved 2026-07-08.
- ↑ "FrontierMath". Epoch AI. Retrieved 2026-07-08.
- ↑ Wiggers, Kyle (2025-01-19). "AI benchmarking organization criticized for waiting to disclose funding from OpenAI". TechCrunch. Retrieved 2026-07-08.
- 1 2 Besiroglu, Tamay; Sevilla, Jaime (2025-01-23). "Clarifying the creation and use of the FrontierMath benchmark". Epoch AI. Retrieved 2026-07-08.
- ↑ "Featured Grants: Epoch AI". Coefficient Giving. Retrieved 2026-07-08.
- promotional language: see Words to watch;
- personal commentary: opinions or direct addresses to the reader;
- informal language.
Instead, only summarize in your own words a range of independent, reliable, published sources that discuss the subject.