Epoch AI — FrontierMath Benchmark Improvements

[Global Catastrophic Risks] Open Philanthropy recommended a grant of $70,000 to Epoch to support improvements to its FrontierMath benchmark for assessing the mathematical reasoning capabilities of AI models. Epoch plans to estimate a human baseline on the test by organizing a tournament for human mathematicians and measuring their performance on a subset of FrontierMath problems. This grant will also support testing to help ensure that the benchmarks fully capture model capabilities. This falls within our focus area of potential risks from advanced artificial intelligence.