It's pretty obviously sarcastic, but it's interesting because ChatGPT thinks it's sincere. It appears that sarcasm detection is still outside the capabilities of large language models, which is surprising because people have been manually annotating sarcastic remarks on internet forums for the past 30 years. How could it get simpler than that to create a training set?
