Timeline for Shall we at least temporarily ban chatbot-generated content (e.g. ChatGPT) until it becomes clearer what it means for Stack Exchange?
Current License: CC BY-SA 4.0
13 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 30, 2023 at 17:08 | comment | added | Rory Alsop Mod | @ErinAnne I think it's still poor quality, so happy to leave it deleted | |
| Aug 30, 2023 at 5:27 | comment | added | Erin Anne | Wondering what to make of this now-deleted answer, which I think contained some truth and some falsehood, one totally spurious attribution, and based on the user's history might have come from Bing? It's better than some of the earlier stuff but still seems to just fall apart under scrutiny | |
| Dec 15, 2022 at 14:41 | comment | added | PM 2Ring | From twitter.com/studentactivism/status/… : "Because if ChatGPT is, as it seems to be, a consummate bullshitter, it's also—definitionally—a bullshitter who doesn't know when its bullshitting. And we all know that that's the most dangerous kind." | |
| Dec 14, 2022 at 22:13 | comment | added | Organic Marble | Noooo! It's an automated BS-unreferenced-answer generator! Kill it with fire. | |
| Dec 14, 2022 at 16:14 | comment | added | called2voyage Mod | @FranckDernoncourt For specialist subjects, I think the odds of it overly copying a single source are fairly high. | |
| Dec 14, 2022 at 15:38 | comment | added | Franck Dernoncourt | @called2voyage makes sense. I'd tend to guess it's quite uncommon that a ChatGPT answer overly copies a single source to the point of requiring some attribution though, given the training set size. | |
| Dec 14, 2022 at 14:58 | comment | added | called2voyage Mod | @Topcode Your first red flag is that an answer has no attribution. While chatGPT can certainly be told to include references, a lot of the chatGPT answers we're getting do not include attribution, and answers with no attribution should be discouraged here anyway. So when people see them, they need to be flagging them for attention. They may turn out to be false positives, but more experienced users can help determine that in the review queues. | |
| Dec 14, 2022 at 14:56 | comment | added | called2voyage Mod | @FranckDernoncourt It's more accurate to say it does not produce valid answers according to the standards of this site. Answers which reference outside content require attribution, and chatGPT answers cannot include guaranteed-accurate attribution. | |
| Dec 14, 2022 at 3:57 | comment | added | Franck Dernoncourt | "It does not produce correct answers". Sometimes it does. | |
| Dec 14, 2022 at 2:56 | comment | added | Topcode | @called2voyage “look like they may have been generated in this way” and can anyone explain what this means? Because there has been at least one false positive on SO already, and in the discussion regarding that nobody has made any progress with creating any sort of guidelines for this. | |
| Dec 13, 2022 at 19:45 | comment | added | called2voyage Mod | @Topcode If we explicitly disallow chatGPT, then it may raise awareness in some portion of casual users to be more careful to avoid upvoting answers that look like they may have been generated in this way. If the answers are less likely to receive upvotes, it's also more likely that the more active users will be able to help us delete these from the review queues without requiring moderator intervention. It's a small thing, but until such point in time that we get a sitewide policy, we'll take what we can get. | |
| Dec 13, 2022 at 18:51 | comment | added | Topcode | “The effort for folks to flag and then mods to track down each of his posts, check and then delete them and suspend the user is not insignificant.” And doing the exact same but because the answer is chatGPT rather than bad, would be less effort? It is possible that some users would be dissuaded by the rules changing, but I think that is an insignificant number. | |
| Dec 13, 2022 at 17:33 | history | answered | Rory AlsopMod | CC BY-SA 4.0 |