Risk Management for Mitigating Benchmark Failure Modes: BenchRisk

Sean McGregor, Victor Lu, Vassil Tashev +8

Large language model (LLM) benchmarks inform LLM use decisions (e.g., "is this LLM safe to deploy for my use case and context?"). However, benchmarks may be rendered unreliable by…