MSSBench

New framework audits safety labels in multimodal LLM benchmarks

Researchers have developed a new framework called the Benchmark Card to analyze the reliability of situational-safety benchmarks used for multimodal large language models (MLLMs). This framework identifies seven distinc…