Snorkel AI co-founder Vincent Chen outlines the company’s expanded Open Benchmarks grants program and its aims for evaluation researchEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria.. He discusses environment complexity, long autonomy horizons and evaluating complex outputs.
Chen explains a proposed collaborative red-team programAdversarial evaluation deliberately searches for inputs and edge cases that reveal where an AI system's rules or behavior fail., argues for continuous benchmark maintenanceA benchmark is a defined test or collection of tasks used to compare a system's performance under specified conditions. and responds to an evaluation audit. He discusses verifier robustness, alignmentAlignment aims to make an AI system's behaviour consistent with intended human goals, values and constraints. and the limitations of relying on a single saturated benchmark.
Watch on YouTube
