Skip to content

benchmark

CommentBench

Benchmark that scores language-model comments on conceptual AI-safety writing by how many of the points human commenters made they reproduce, using an automated matcher.

Current clusters