AI benchmarks
•
Applied to Trendlines in AIxBio evals ago
•
•
Applied to XPT forecasts on (some) Direct Approach model inputs ago
•
Applied to AI Forecasting Research Ideas ago
•
Applied to Results from an Adversarial Collaboration on AI Risk (FRI) ago
•
Applied to Long list of AI questions ago
•
Applied to Language models surprised us ago
•
Applied to $250K in Prizes: SafeBench Competition Announcement
ago
•
Applied to Open Phil releases RFPs on LLM Benchmarks and Forecasting ago
•
Applied to Encultured AI, Part 1: Enabling New Benchmarks ago
•
Applied to Race to the Top: Benchmarks for AI Safety ago
Benchmarks are tests which enable us to measure the progress of AI capabilities, and test for characteristics which might pose safety risks.