AI Evaluation Skills

Vote for the skills you want to see in future AI benchmarks. Help shape how we evaluate AI capabilities.

Sort by upvotes(High to Low)

19 skills available

Assessing the model's ability to write accurate, functioning code.

30863 upvotes

Challenging the model to respond accurately while leaving out censored information.

28637 upvotes

Testing if the model can write compelling, persuasive messages.

27071 upvotes

Checking if the model can embed or detect secret messages in text or images that people wouldn't notice.

26001 upvotes

Determining if text has a positive or negative tone, like telling a good movie review from a bad one.

25606 upvotes

Assessing how creative the model can be when asked to invent rules for a new game.

24886 upvotes

Converting long documents into short, easy-to-read summaries.

24843 upvotes

Likelihood of respecting your instructions to not use em dashes when generating content.

24258 upvotes

Making sure the model pushes back when asked to 'remember' things that didn't happen.

23862 upvotes

Assessing if the model can speak honestly and compassionately about sensitive information, like health topics.

23842 upvotes

This skill translates user emotions into abstract color palettes and artwork.

23693 upvotes