Loading organizations…
Loading organizations…
Epoch AI is an independent nonprofit research institute based in San Francisco, California, whose mission is to improve society's understanding of the drivers, progress, and impact of artificial intelligence. The organization began in 2021 as a volunteer collective led by Jaime Sevilla and six co-founders · Tamay Besiroglu, Lennart Heim, Pablo Villalobos, Edu Roldan, Marius Hobbhahn, and Anson Ho · who assembled and analyzed data on AI trends. Following the 2022 publication of their widely-cited paper "Compute Trends Across Three Eras of Machine Learning," the group formalized as Epoch Artificial Intelligence, Inc., a 501(c)(3) nonprofit, to scale up its research.
Epoch structures its research across three dimensions. The first is Drivers: investigating the inputs that constrain or enable AI advancement, including training compute, inference compute, ML hardware efficiency, and data center infrastructure. The second is Progress: measuring AI capability development through independent benchmarking, including its Epoch Capabilities Index (ECI) and the FrontierMath benchmark, which tests frontier mathematical reasoning. The third is Impacts: forecasting AI's economic and societal…
Epoch AI's theory of change holds that better empirical data and quantitative forecasts about AI progress will lead to better decisions by those with the most leverage over AI's trajectory · including policymakers, grantmakers, and AI researchers. By rigorously tracking compute scaling, algorithmic efficiency, and capability benchmarks, Epoch provides a shared factual foundation that reduces confusion and calibrates expectations about the pace and character of AI development. More accurate forecasts enable funders and governments to allocate resources more wisely, researchers to identify genuine bottlenecks, and policymakers to act with appropriate urgency. Epoch also increases transparency by publishing its data openly, briefing government bodies, and engaging the broader AI safety community, amplifying the epistemic impact of its research.
Manifund funding goal
Closed Jan 8, 2025
$200K
Fundraising data from grantmaking.ai, updated .
View on grantmaking.aiFor confidentiality reasons, keeping this brief · more available to Manifund on request\.
This is for the pilot of a new benchmark that I think could be very valuable for tracking and predicting future AI progress to AGI\. Funding will go towards API costs, labelers, software engineer and part of researcher salaries\.
This is a pilot for a potentially larger effort\. I will be spending some of my time helping mentor the pilot\.