Prompt Refusal
Data Skeptic
00:00
How to Classify Chat GBT Responses Automatically
The data set we used was from the public question and answer site Quora. We found that it had a pretty good mix of normal questions and edgy or offensive or insulting text strings that chat GBT was likely to refuse. The prompt classifier would take an impromptu, tell you whether it was likely to be accepted or rejected. And if chat GBT doesn't have a clear imperative or a question, it's really hard to make the judgment on whether it's refusing the prompt or not.
Transcript
Play full episode
