←back to thread

AI agent benchmarks are broken

(ddkang.substack.com)
181 points neehao | 1 comments | | HN request time: 0.409s | source
1. btdmaster ◴[] No.44537136[source]
There is a cool solution for this: https://huggingface.co/spaces/Jellyfish042/UncheatableEval

This doesn't work for instruction-tuned models, but it's an interesting alternative approach that doesn't need a complicated (and thus gameable) evaluation function or human interaction. Instead, predict the next word with data newer than the training set.