News for hackers. Stuff that matters.
Re:ChatGPT has trouble giving an answer before explaining it... by ugurnot on Tuesday March 07, 2023 @09:09PM (#35064301)
I evaluated ChatGPT on Winogrande Debiased validation set[1], a dataset focused on commonsense reasoning. ChatGPT has an accuracy of 62.75%, below GPT-3's reported accuracy of 77.7%.

https://github.com/ugorsahin/Winogrande_ChatGPT

There are no comments on this comment yet.