Spaces:

CoreyMorris

/

MMLU-by-task-Leaderboard

like 13

Running

I noticed some odd things from the Hugging Face Open LLM Leaderboard MMLU results. After investigating further, I found that variations in performance on the moral scenarios task were not caused by differences in the ability to comprehend and apply good moral judgement, but rather by the answer choices' structure.
https://medium.com/@coreymorrisdata/is-it-really-about-morality-74fd6e512521

• Remove or modify section related to MMLU's moral scenario task