[eggsbenedict]

This implementation produced almost no useful answers. One of the 'Friendly No' outputs contained this sentence:

"No, I'm sorry I can't attend the meeting because the topic is ridiculous"

Which is about as unfriendly as it gets.

[Impossible]

This app is a perfect example of why you shouldn't show "raw" GPT-X output to users. "No" produced some great stuff like " No, I must not attend the meeting because the children cannot witness its nature." The site works as a machine learning joke, but its not clear if that's what the creator is going for...

Edit: I read the disclaimer at the bottom, there is some filtering (which explains why I was getting only casually racist, violent and anti-muslim answers and not blatantly offensive content :)). Also it's a one day hack project, so it is basically an ML joke.