Timeline for Policy: Generative AI (e.g., ChatGPT) is banned
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 25, 2023 at 8:32 | comment | added | Cerbrus | I don't have any numbers, but multiple mods have stated that question range didn't drop at all, @joesan | |
| Jan 25, 2023 at 5:32 | comment | added | joesan | Is there some data to measure the question rate? I mean after chatGPT was launched, did SO notice a drop in the question rate? Are there some numbers that compares pre and post chatGPT release having an effect on the question rate per minute here at SO? | |
| Dec 23, 2022 at 16:28 | comment | added | Summer-Sky | While language models cannot provide all of the things that are demanded of them, they can still be an assistive technology in two ways: by summarizing long text for easier ingestion or by helping to write it in a well-structured manner. | |
| Dec 8, 2022 at 13:21 | comment | added | Zoe - Save the data dump Mod |
Piling on examples of CGPT producing garbage, I tried it in this question over on Vi.SE, where CGPT recommended to change font to guifont (which isn't a valid property in that context). When I told it that, it said to use font, and when I told it that doesn't work, right back to guifont. (I've more or less established that particular problem as a bug in Vim though, so it producing anything useful was already extremely unlikely. Was for science, and to see what it would answer, but makes for a great example)
|
|
| Dec 8, 2022 at 13:16 | comment | added | Cerbrus | It can, but there's no guarantee. Nowhere am I claiming that writing a good question guarantees a correct answer. It's just more likely. | |
| Dec 8, 2022 at 13:01 | comment | added | Erik A | Your answer makes it sound like it can answer clear, well written questions that could get answered by humans. My point is: that's only true if the answer is trivial too, since cGPT fails to solve clear and short questions with slightly difficult answers that humans can answer just fine. It even fails to interpret them and changes them to something that it has an answer to. | |
| Dec 8, 2022 at 12:36 | comment | added | Cerbrus | That's basically my point :D | |
| Dec 8, 2022 at 12:36 | comment | added | Erik A | It's currently 0/8 out of the reasonably short questions I've asked, so if that's the most likely, well, that's not great. And the answers it does come up with fall either into "plain wrong code", "an imaginary fun/arg that solves your problem but doesn't exist" or "you've missed the entire point of the question". I don't think CGPT can answer anything but the most trivial questions, even when the question is well formulated. And those trivial questions are likely duplicates, of course. | |
| Dec 8, 2022 at 12:16 | comment | added | Cerbrus | @ErikA: "They're the most likely to get good output" That doesn't necessarily mean they will be good :D | |
| Dec 8, 2022 at 11:44 | comment | added | Erik A | I've arrogantly tried it on a couple of my own questions assuming those are good, but it failed miserably. E.g. this one, a VBA terminate handler not firing due to a reference loop between the class and some built-ins, it suggested "move the code to the initialize handler, then it'll fire" 🤦♂️. When asking for unicode normalization for password storage, it provides an example that mojibakes the string... | |
| Dec 8, 2022 at 9:48 | history | answered | Cerbrus | CC BY-SA 4.0 |