Timeline for answer to Policy: Generative AI (e.g., ChatGPT) is banned by Rowanto
Current License: CC BY-SA 4.0
Post Revisions
22 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Feb 25, 2025 at 9:33 | comment | added | CharonX_has_given_up_on_SE | Rowanto was truly ahead of their time. SE is now running an experiment with LLM generated answers to questions on some sites, asking users to "review" those AI answers... and it is exactly the awful inane LLM crud we warned against / expected it to be. | |
| Apr 4, 2024 at 23:11 | comment | added | CPlus Says Cancel The Redesign | Do we really need a new 'ChatGPT Answers' review queue? | |
| Nov 1, 2023 at 13:44 | comment | added | Security Hound | "The problem is now that a user can't really differentiate them." - It's trivial to identify ChatGPT generated content. The only way to handle ChatGPT content is to ban it. | |
| Sep 28, 2023 at 6:39 | comment | added | MisterMiyagi | @Xartec "which would discourage answerers from using the same ai to produce sub par results" Why? How? The answer also just takes this magic for granted, so please provide arguments for it if you think it is actually true. I’ve seen many AI-spammers post indiscriminately on questions that already had answers, so it is not at all clear to me why the existence of official AI answers would prevent or discourage AI spam. | |
| Sep 27, 2023 at 21:34 | comment | added | user400654 | There is no magic prompt engineering masterpiece that will suddenly make any current or near release version of gpt capable of producing answers to new questions. At best it can summarize an existing answer to an already answered question, however even that becomes more likely to hallucinate than to provide an accurate answer the more you allow it to modify the content it is summarizing to fit the person looking for an answer. The user would be better off just getting the existing answer. | |
| Sep 27, 2023 at 21:19 | comment | added | user400654 | @Xartec Yes, and we've been routinely deleting said answers for nearly a year now. Yes, we haven't gotten all of them, but we've gotten enough, and enough accounts are actioned against to keep the problem at bay. Your assertion that we cannot discern whether or not content was written by AI is quite provably false. | |
| Sep 27, 2023 at 21:16 | comment | added | Xartec | @KevinB the why has been established, the problem at hand and is repeated in my previous comment. In an ideal world you’d be correct, that would surely be the best person. In reality, people post answers generated by AI regardless of whether SO ‘must’ and no the person with the problem is surely not automatically the most capable of asking the question. That assumes everyone masters prompt engineering which just isn’t true. We can disagree on opinions, not facts. An imo, a more controlled AI answer would likely reduce the urge to create a poor Ai generated answer. | |
| Sep 27, 2023 at 21:06 | comment | added | user400654 | @Xartec they've already tried that, it failed spectacularly. Many other sites are similarly trying this, such as quora, and are having similarly questionable results. Why must SO include AI generated answers? What value would that provide? Surely the best person to have a conversation with a chatbot and weed out poor results is the person with the problem being solved, not some rep hunter on SO or SO itself. | |
| Sep 27, 2023 at 21:03 | comment | added | Xartec | @Cerberus “ Why do we need to invest time in implementing generated crappy answers, when users can get that nonsense at the source, instead?” This, again, relies solely on the false idea AI generated answers are crappy and nonsense. Which is is like building a house on quicksand, as chat gpt would say. The quality of the answers can be heavily improved by prompt engineering which would discourage answerers from using the same ai to produce sub par results. It would be a sensible way to introduce AI, an inevitable situation. | |
| Sep 27, 2023 at 20:57 | comment | added | Xartec | @MisterMiyagi that’s what his proposed solution is about. His answer is a solution adressing the problem as layed out in the second sentence. | |
| Sep 27, 2023 at 13:24 | comment | added | MisterMiyagi | @Xartec "It’s however about answerers using AI" To the contrary. This answer is about SO itself using AI, which then - somehow - removes the issue/occurrence of answerers using AI and - somehow - offers a benefit to askers. | |
| Sep 27, 2023 at 13:13 | comment | added | Cerbrus | @Xartec Rowanto is arguing that SO should implement GPT to automate answers. That's what my "Why" is asking about. Why do we need to invest time in implementing generated crappy answers, when users can get that nonsense at the source, instead? | |
| Sep 27, 2023 at 12:35 | comment | added | Xartec | @Cerbrus, the second question in your comment isn’t a relevant as it may seem. You ask why something can’, while of course an asker can and is not prevented from doing so regardless. Just as the asker could rtfm or get an answer elsewhere. It’s however about answerers using AI, who do that regardless of what the asker could do. | |
| Sep 21, 2023 at 0:34 | comment | added | Félix Adriyel Gagnon-Grenier | Like let's see how good exactly can we train a model exclusively on SO content to answer duplicates. | |
| Sep 21, 2023 at 0:34 | comment | added | Félix Adriyel Gagnon-Grenier | I must admit I do not actually dislike this idea as much as most other suggestions at face-value. I mean it's not what I would like, but if a compromise had to be struck, in the event where we cannot absolutely reliably ban ai generated answers, I would choose a somewhat "site-approved" or even better, "site-trained" ai answer generator. | |
| Sep 20, 2023 at 21:12 | comment | added | Peter Mortensen | cont' - Sample code dump answer, 1. Sample code dump answer, 2. Essentially a sample code dump answer, 3. | |
| Sep 20, 2023 at 21:03 | comment | added | Peter Mortensen | Re "not yet checked by a human": This doesn't (and didn't) even happen with human-generated answers to any significant extend. Why would anyone spend time checking a code dump answer (no explanation whatsoever)? Yes, that is a rhetorical question. A code dump answer may be completely bogus or brilliant. It is difficult to judge unless significant time is spend to actual run and test the code. | |
| Sep 20, 2023 at 21:00 | history | edited | Peter Mortensen | CC BY-SA 4.0 |
Active reading.
|
| Sep 19, 2023 at 20:49 | history | edited | Heretic Monkey | CC BY-SA 4.0 |
name of the site is two words
|
| Sep 19, 2023 at 19:52 | comment | added | MisterMiyagi | If the AI reliably produces correct answers - why build a library of them at all? If the AI does not reliably produce correct answers - how many and which answers should be posted? Critically, how should volunteers deal with the required massive content volume when we already have too few people for the fewer human generated answers? | |
| Sep 19, 2023 at 13:52 | comment | added | Cerbrus | I think you're massively underestimating the effort that goes into training this kind of model... Also, why does the generated answer need te be on SO? Why can't a user that wants an AI-generated answer just go to said AI? | |
| Sep 19, 2023 at 13:37 | history | answered | Rowanto | CC BY-SA 4.0 |