3 ms·
This case is slightly weird. Without evidence, there's no real way to call it out without sounding hostile, which is something you've also discouraged on Hacker
by minimaxir 6y ago
This case is slightly weird. Without evidence, there's no real way to call it out without sounding hostile, which is something you've also discouraged on Hacker News.
It's also why adequate disclosure of AI-generated text is important and it's good OpenAI is advocating that.
- dang 6y agoI don't think https://news.ycombinator.com/item?id=23894000 https://news.ycombinator.com/item?id=23894000 sounded hostile. I agree that it's an edge case. I don't think moderators would have replied to https://news.ycombinator.com/item?id=23894742 https://news.ycombinator.com/item?id=23894742 as a guidelines violation, but the user who did meant well and all this stuff is pattern-matching anyhow. For example "this sounds like GPT-3" can just as easily be an internet insult (e.g. "sounds like GPT-3 regurgitated a sociology textbook" - https://news.ycombinator.com/item?id=23687199 https://news.ycombinator.com/item?id=23687199), in the same way that "this sounds like a Markov chain" used to be. The fact that that comment added "or the human equivalent" kind of made it lean that way. Perhaps that's why it provoked an immune response from a community member while the more neutral "I think this was written by GPT-3" comment did not.