Wikipedia:Reference desk/Computing

Welcome to the computing section
of the Wikipedia reference desk.

skip to bottom

Select a section:

• Computing
• Entertainment
• Humanities
• Language
• Mathematics
• Science
• Miscellaneous
• Archives

Want a faster answer?

Main page: Help searching Wikipedia

   

How can I get my question answered?

• Select the section of the desk that best fits the general topic of your question (see the navigation column to the right).
• Post your question to only one section, providing a short header that gives the topic of your question.
• Type '~~~~' (that is, four tilde characters) at the end – this signs and dates your contribution so we know who wrote what and when.
• Don't post personal contact information – it will be removed. Any answers will be provided here.
• Please be as specific as possible, and include all relevant context – the usefulness of answers may depend on the context.
• Note:
  • We don't answer (and may remove) questions that require medical diagnosis or legal advice.
  • We don't answer requests for opinions, predictions or debate.
  • We don't do your homework for you, though we'll help you past the stuck point.
  • We don't conduct original research or provide a free source of ideas, but we'll help you find information you need.

Ready? Ask a new question!

How do I answer a question?

Main page: Wikipedia:Reference desk/Guidelines

• The best answers address the question directly, and back up facts with wikilinks and links to sources. Do not edit others' comments and do not give any medical or legal advice.

See also:

• Teahouse
• Help desk
• Village pump
• Help manual

July 27

Accessing "old" gen AI response before a Wikipedia article was published?

Odd question here - I'm a high school teacher, and created the Wikipedia page Capitol Crawl to use as a demonstration for my students (also because it well exceeds the notability guidelines and should've had its own page ages ago!). Many popular websites say or imply that the Capitol Crawl was a major catalyst for the passage of the ADA, but I found in my research that the reality is more nuanced, so the page reflects that.

When I ask Chat GPT about the impact of the Capitol Crawl, its response closely mirrors what I wrote, and cites the Wikipedia page as its source. I'm confident that these responses would have been different before I created the article, and I'd love to use this as an example for my students of the limits of AI and how AI models get their information. But sadly, I didn't think of it until now! Is there a way I can see what an AI model like Chat GPT would have said about this topic before the article was written? If it's relevant, the article was created on May 10th. MissRedwood (talk) 22:52, 27 July 2026 (UTC)

If you ask AI about the impact of the Capitol Crawl and its answer is very similar to what you wrote on Wikipedia, you can then ask a follow-up question like "Ignore Wikipedia and answer the same question." Stanleykswong (talk) 23:08, 27 July 2026 (UTC)

And then they can politely say, "Yes, OK, of course", and proceed to answer by using info they got from Wikipedia during the learning phase, distributed in an inextricable way across the possibly trillions of parameters (weights) of the neural network.  ​‑‑Lambiam 04:23, 28 July 2026 (UTC)

I've never tried this method on Wikipedia, but I always use "ignore sources from China and answer the same question" to get more accurate economic information about China. Stanleykswong (talk) 05:21, 28 July 2026 (UTC)

What Lambiam is saying is that when the LLM is trained it uses billions of pages of text to adjust trillions of parameters (numbers). Text that it's given changes these values to improve how the model writes text. It doesn't store the text but just adjusts the numbers, similar to how a book I read five years ago is going to adjust how I communicate -- I'm not remembering the book word for word, it's just somehow influenced me in some way, and it can be extremely difficult knowing how that book affected my brain. If I told myself to ignore that book, I would not do a good job adjusting what I write because I have no idea how that book influenced me. Panamitsu (talk) 06:27, 28 July 2026 (UTC)

An article you just created yesterday cannot be included in the learning phase without time travel. In solidarity, Aaron Liu (talk) 15:06, 28 July 2026 (UTC)

Is it May 11 where you are? Matt Deres (talk) 13:30, 29 July 2026 (UTC)

The OpenAI training cut-off date is December 2025. In solidarity, Aaron Liu (talk) 14:21, 29 July 2026 (UTC)

If that is the case, then the chatbot's reply with similarity to the new Wikipedia content is not from training it into the weights, but from the chatbot using the web - so telling it to not use the web should work. Icek~enwiki (talk) 13:32, 2 August 2026 (UTC)

This question has already been asked at the Help desk where it has multiple responses. Shantavira|^{feed me} 07:08, 28 July 2026 (UTC)

July 30

Where to Report Two Incidents

I would like advice on where to report two (unrelated) software incidents (so probably two different places).

About two weeks ago, I was trying to make an archival copy of my C: drive to a large external HDD, using ROBOCOPY. I forgot to specify the /XJ switch on the ROBOCOPY, which resulted in about 1 TB of data being copied (from a C: drive that had about 230 GB of data) to the archival drive before I killed the copy. I know what that problem was, which was a ROBOCOPY runaway in following symbolic links. Then I tried to delete the recursive archival data from the archival drive. After some effort, I tried one last effort, and can look up exactly what I did, but am not looking it up yet. My last attempt to delete the recursive data resulted in no response after an extended period of time, after which I made various attempts to interact with my computer, and I eventually had to unplug the computer from the UPS and restart it. Analysis of the Event Viewer log showed that memory had been exhausted. I also lost one of the directories on my C: drive. The attempt to delete recursive data had caused a loss of real data. My conclusion is that I shouldn't have tried to delete the recursive directory that had been created by an out-of-control process, and that formatting the external disk was the way to minimize the damage. So my question is what is a good forum in which to describe this incident and the lesson learned.

The second incident was that I was trying to use Microsoft Copilot to help me walk through the process of replacing the cartridge for an HP 139W laser printer. It told me to open the front door of the printer. I said that I had opened the top cover of the printer and could access the cartridge, and it told me not to do that, but that I should open the front door instead, and that changing the cartridge through the top could result in damage to the cartridge. There isn't a front door, only a top cover. Other printers presumably have a front door. How can I report to Microsoft that Copilot was hallucinating about the HP 139W? What is a good forum to advise users of Microsoft Copilot that it may hallucinate about the design of a printer?

These are two matters that are unrelated except that I would like to know a good forum to describe each incident for the benefit of other users. Robert McClenon (talk) 02:26, 30 July 2026 (UTC)

Based on your description, ROBOCOPY behaved as it should. You didn't include /XJ to tell it to skip symlinks. What would you report to ROBOCOPY? You made a mistake and you deleted files you shouldn't have deleted after that?

As for Copilot, it is an LLM chatbot. They are not intelligent. They only put sentences together that appear to make sense. They all come with legal disclaimers that you need to verify the correctness of the LLM's output. ~2026-41824-87 (talk) 11:25, 31 July 2026 (UTC)

A video shows opening the top cover of the laser printer that the OP may recognise https://www.youtube.com/watch?v=2WrUspm9tqg . When I ask CoPilot for instruction it replies "here are the exact, official steps for replacing the toner cartridge in the HP LaserJet M139w/M139we series, based directly on HP's support documentation and videos." and it provides links to HP Support instructions and a replacement video overview. CoPilot arguably covers the OP's issue about whether one opens the top or opens the front by quoting "01 Open the toner access door....Lift or open the front toner access door fully." I understand CoPilot has no image interpretation ability to distinguish between a door and a cover. CoPilot comprehends them both as "access door" which is perhaps forgiveable. However to warn that changing the cartridge through the top is forbidden and damaging is bad advice. My experience of correcting CoPilot's mistakes is that when confronted it readily "apologises for the confusion" but neither Microsoft nor CoPilot will pretend never to make a similar mistake again. ~2026-39925-22 (talk) 20:46, 2 August 2026 (UTC)

August 3

Unblock my simcard puk code

Give me the puk code simcard +966545876963 ~2026-43097-65 (talk) 18:17, 3 August 2026 (UTC)

@~2026-43097-65 Nobody knows your PUK code except your mobile service provider. It may be available on the SIM card's packaging, the contract, or provided by their customer service after identity verification. You will need to call them. Shantavira|^{feed me} 07:11, 4 August 2026 (UTC)

August 4

Chromium renders visual editor using 2GB RAM (on large pages)

I presume this was not part of the design... and for that I have some serious debugging work to do.

debug_info:
MacBook Air | M1 - 8GB - macOS 26.5.2
ungoogled-chromium | arm64 - 144.0.7559.96
Display enlargement 200%
User-defined font: IBM Plex Sans (variable, 16 pt)

海盐沙冰 _{/ aka irisChronomia / Talk} 09:13, 4 August 2026 (UTC)

Is THAT why it was bogging down so badly the other day?? I had to help someone on Teahouse with a workaround to disable it because it was locking up the entire page - couldn't even load in to swap over to source mode. Phones are not made for that ish! (Me and Chromium's insatiable appetite for memory have had issues for a hot minute.) ChompyTheGogoat (talk) 11:03, 4 August 2026 (UTC)