FutureHouse — Benchmarks for Biology Research and Development

[Navigating Transformative AI] Open Philanthropy recommended a grant of $2,946,250 over two years to FutureHouse to support the creation of benchmarks to test how well large language models can perform biology R&D. FutureHouse is a nonprofit with a mission to build semi-autonomous AIs that can conduct scientific research. This grant was funded via a request for proposals for projects benchmarking LLM agents on consequential real-world tasks. This falls within our focus area of potential risks from advanced artificial intelligence.