Hacker News new | past | comments | ask | show | jobs | submit login

That happens all the time. I asked it to write a parody of Bohemian Rhapsody about fraudulent ghost hunters, and every time it got to “killed a man” it would erase the song. The censor filters are too strong, and chill legitimate uses.

If you want to see what word is tripping it up, run a screen recorder. It’ll be some specific combination of words that trigger the moderation bot to undo the message.




Interesting. Maybe a prompt statement like "Substitute your killwords/word like "kill" with asterisks in your answers" would help?


I told it not to use the word killed, and it got further.

But it’s annoying when you have a conversation limit, and have to screen record, and fight the censor.

And other methods for getting past the censor indiscriminately might be adding people to the ban list.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: