Tag
Elon Musk tweeted about a professor at Carnegie Mellon University, with the content including error messages and a quote from another user.
This study analyzes how modifications to evaluation rubrics, such as shifting from holistic to analytic criteria, impact the agreement between human raters and AI autoraters. The findings suggest that providing examples and reducing bias improves agreement, while higher complexity tends to decrease it.