Wikipedia:Bots/Noticeboard

Basically the title. I tried searching the archives here and on similar pages for AI and LLM discussions but nothing seemed to jump out. What I mean is: not calling or invoking LLM with a program to decide on conent.

I mean: using LLM to create deterministic code that would be a bot to edit Wikipedia, but to develop it WITH LLM/AI tools? For instance, using Claude Code or Codex to generate something that would be Github accessible, and run on a system that I control to edit Wikipedia in some approved bot manner. But, all the "programming" of the bot code would be done by LLM/AI.

Is that any WP:NOLLM concern? Why is it or is it not different? — VPP (t/c) 00:08, 13 October 2026 (UTC)Reply

One can test LLM-generated code in exactly the same ways one should test non-LLM generated code, and once one has verified that LLM-generated code functions as expected then it will consistently function that way every time. Programmers already expect to having to rigorously test their output and not trust superficially correct code, in a way that we absolutely do not for written text. Most of the issues that prompt NOLLM (hallucinations, text-source integrity, bias, weird writing style) do not apply in the same way to code. I can imagine a policy against LLM-generated code (though I'd oppose such a policy without evidence that we're seeing issues caused by this), but it would need to be considered separately since it's just a completely different beast with very different classes of potential issues.

More cynically, the people writing code with LLMs onwiki have, on average, vastly more clue than the people writing articles with it, and so are less likely to use it in stupid ways or release inadequately checked LLM output. If we had even a fraction of the volume of codeslop in bots as we do in articles we'd absolutely need some way to regulate it. Rusalkii (talk) 00:51, 13 October 2026 (UTC)Reply

There's currently no rule about vibe-coded bots, and it would be difficult to enforce one since we don't require that bot code be open source, and when it is I don't think we often bother to actually review it. Instead we generally judge bots based on behavior during the trial and afterwards: if the bot is making too many errors, it'll not be approved, or if already approved it'll be unapproved and blocked if necessary. And for what it's worth, it's pretty likely that many of the non-AI-coded bots aren't written to any sort of professional standards either. Also for what it's worth, "professional standards" these days seem to be all over the place when it comes to AI-written code; it's not uncommon to see supposed professionals slopping AI slop all they can, and then bragging about how "productive" they're being by becoming "managers" of a bunch of AI agents that test and review each others' code, while the people-managers cheer them on as they hope to be able to replace the human managers-of-teams-of-AI-agents with another layer of AI. Anomie⚔ 02:36, 13 October 2026 (UTC)Reply

You asked why the difference. As already mentioned, the trial run process helps reduce errors. As an additional reason we don’t allow LLM generated text in articles is the licensing is disputed. That’s not an issue for bot source code because we don’t require it’s published under any particular license. Dw31415 (talk) 02:55, 13 October 2026 (UTC)Reply

Last I checked every programmer I knew, including the ones with long opinions about how AI is the devil, has been making extensive use of the machine that writes good code. It's certainly possible that the community consensus could end up being that you aren't allowed to use Claude or Codex, or that you aren't allowed to use an IDE with tab-complete, or that you are required to write all your code by starting out with echo "#include <stdio.h>\nint main() {" >> program.c — I just don't think it is very likely (or at least I hope it's not). jp×g🗯️ 21:33, 1 November 2026 (UTC)Reply

Well, now you know of one who doesn't. Anomie⚔ 21:50, 1 November 2026 (UTC)Reply

Now you know of two. * Pppery * _{it has begun...} 00:12, 2 November 2026 (UTC)Reply

For all the code I write for Wikipedia I use an LLM to write comments (everyone knows Python and JS but I mostly work with JS or R), format the code so it's pretty, create barebones docs (because I'm lazy), double check my work, and to speed up debugging problems. But I write all the rest of my code by hand or just reuse code from older projects. If my code is slop it's the finest human slop or reused decade old code I wrote in grad school. I try to put a little blurb in the code comments or readme that says which AI model I used and what I used it for, keyword is try. I've been also trying to post more of my code to a dedicated GitLab profile for just Wikipedia things. Dr vulpes (Talk) 01:44, 2 November 2026 (UTC)Reply

See . All bots running on Toolforge are down. See also WP:VPT#Preparing for a short-notice security update to our services. Tenshi! (Talk page) 23:53, 20 October 2026 (UTC)Reply

Moved from Wikipedia:Bots/Requests for approval/ZackBot 14 (permalink)

{{BAG assistance needed}} A request has been made for me to apply for a specific exemption to the generally-accepted norm of 20 edits per minute that is specified in WP:BOTPERF. The current cleanup of Template:Infobox football biography has been causing some issues of epople trying to clean it up on their own and not doing it correctly. Jonesey95 has specifically request that I expedite this cleanup due to the issues (Jonesey please chime in here). Would be great to get this exception in place... --Zackmann (^{Talk to me}/_{What I been doing}) 21:08, 24 October 2026 (UTC)Reply

I think 20 edits per minute is fine. There are 200,000 articles, so that would be 7 days of continuous editing. It will take longer for this bot, because it can't run 24/7. But that's okay. This is a voluntary, non-urgent task to standardize parameter names per MOS. The risk in running the bot more slowly is that editors are seeing the "deprecated parameters" category and error message and make manual changes, which they sometimes do incorrectly. That will require some manual cleanup after the bot is done running. It's a downside, but not a huge one if the community believes that the 20-edit-per-minute limit is an important one to enforce. – Jonesey95 (talk) 21:44, 24 October 2026 (UTC)Reply

I guess I just don't understand this....

1. This is an approved task
2. The account has the bot flag.
3. This specific cleanup has been extensively discussed
4. The bot has over 700k edits and has been well tested

What is the virtue of hand tying the cleanup and restricting the speed here? I am not saying that this is some super urgent, critical task that needs to be done immediately or the Wiki will collapse... But what is the goal of making this run slower than it can? What is the benefit? @Primefac: can you shed some light here? What am I missing... Zackmann (^{Talk to me}/_{What I been doing}) 22:04, 24 October 2026 (UTC)Reply

The usual reason is that if, for some reason, the bot started to have a bug, or the relevant infobox changed midrun, or something like that, having the bot pace itself makes it easier for humans to a) review the changes, b) stop the bot at e.g. 2,000 edits instead of 175,000 edits. In practice, that usually means respecting mw:Manual:Maxlag_parameter for anything non-urgent. Which this isn't. Headbomb {t · c · p · b} 23:30, 24 October 2026 (UTC)Reply

@Headbomb: appreciate the explanation. I guess my point is if the bot has gone through a BRFA and has performed over a half million edits, it should be trusted already or it should not be an approved bot... The point of a bot is to get things done. I guess I just take a different stance that hand-tying the process so that each edit can be reviewed, defeats the entire point of a BRFA... Zackmann (^{Talk to me}/_{What I been doing}) 23:44, 24 October 2026 (UTC)Reply

Not saying I agree/disagree with the limits in this instance, just that this is usually why the limit is in place. A well-tested bot has a lower potential for suddenly shitting the bed, but an API change is always a possibility, or infobox template change, etc... 20/min subject to maxlag seems a very reasonable limit to me. Like others said, that would clear the backlog in ~7 ish days. Headbomb {t · c · p · b} 23:47, 24 October 2026 (UTC)Reply

You make good points. I will say that the API changing was not a possibility I had considered. I also want to be clear that the 7ish days is CONTINUOUS editing, which my bot is not setup for. It only runs when my browser is open and running. So realistically we are looking at 3-4 weeks at best. Does that warrant unrestricted editing rate? Not sure, but just want to be clear about the timeline here. Zackmann (^{Talk to me}/_{What I been doing}) 00:57, 25 October 2026 (UTC)Reply

One, at the bottom of an already-approved BRFA isn't really the place for this sort of discussion. It should probably have been started at WP:BON. Two, besides the "hopefully catch malfunctions before they make a huge mess" thing Headbomb mentioned, the "20 edits per minute" rule also serves to reduce flooding of watchlists and such. Overall, it doesn't sound like "some humans might make a few bad edits" is that much of a worry to bypass the rule; even Jonesey95 doesn't seem to think it's that big a deal. Nor does "this will take weeks"; that's the nature of things sometimes, and for example User:CSD U6 Bot will probably be working on its backlog for another 4 years since consensus is that it should be limited to 150 backlogged pages per day. If your bot not being able to run continuously is an issue for you, you might look into reworking it for a non-browser environment (e.g. nodejs) and running it on Toolforge. Anomie⚔ 04:01, 25 October 2026 (UTC)Reply

@Anomie: I wasn't sure where to start the discussion. Since I was asking for a addendum to a specific BRFA, this felt like the best venue. In the future, I'll def go to WP:BON instead.

Re the flooding of watchlists, isn't that the point of the bot flag? So that all these edits can be ignored on the watchlist?

Your note about finding a way to run it continuously is 100% valid and already something I am looking into.

I want to be clear again that I don't mean to suggest or imply that this bot is somehow so critical that it needs to be run ASAP or else. I'm probably just impatient and want to get this done so we can move on to the next task.

Appreciate all the insight and advise on this issue. Zackmann (^{Talk to me}/_{What I been doing}) 04:07, 25 October 2026 (UTC)Reply

I note the header and footer of the close box both say To request review of this BRFA, please start a new section at Wikipedia:Bots/Noticeboard. The other common way to request an "addendum" is to file a new BRFA (cf. this discussion).

While the bot flag helps reduce the flooding, limits are still appropriate. Anomie⚔ 04:27, 25 October 2026 (UTC)Reply

(BAG Note: I moved this from the BRFA per multiple comments)

I largely concur with my colleagues above; this does not seem like such an urgent task that it needs to exceed the usual rate limit of 20epm. I don't think I have much more to add on the matter but happy to do so if I need to address any specific concerns. Primefac (talk) 21:52, 26 October 2026 (UTC)Reply

How do I get the Cewbot to update this? Hawkeye7 (discuss) 23:24, 27 October 2026 (UTC)Reply

Doesn't the bot already update it? Primefac (talk) 09:18, 1 November 2026 (UTC)Reply

• Rick Bot (BRFA · contribs · actions log · block log · flag log · user rights)
• Wikipedia:List of administrator hopefuls (edit | talk | history | links | watch | logs)

Following Wikipedia:Categories for discussion/Log/2026 October 19 § Category:Wikipedia administrator hopefuls, Category:Wikipedia administrator hopefuls was deleted by consensus. This category was used by Rick Bot task 4 to update Wikipedia:List of administrator hopefuls, which has now been marked as {{historical}}. However, Rick Bot is still running task 4. I am filing this request as a pro forma WP:BOTAPPEAL to revoke the approval of task 4, given it no longer enjoys consensus.

I have notified the botop, Rick Block, of this filing. I have not attempted to discuss with them because they have not edited in seven months. Best, HouseBlaster (talk • edits • he/they) 00:14, 30 October 2026 (UTC)Reply

Revoking the approval won't stop the bot, you'll need to get the operator to respond for that. Come back if the operator refuses to stop the task for some reason. Anomie⚔ 02:15, 30 October 2026 (UTC)Reply

Given that the bot is actively malfunctioning and the bot operator has not edited in seven months, I suspect the bot will need to be (at least temporarily) pblocked. I didn't believe that requesting a pblock before the task was revoked made sense, but I'm happy to be corrected on that front. Best, HouseBlaster (talk • edits • he/they) 02:24, 30 October 2026 (UTC)Reply

"Actively malfunctioning" is a stretch, IMO, for a category that you just had deleted, affecting exactly one page that, by your own admission in the deletion discussion, no one cares about. I tell you again: try to contact the bot op first. Try email if they don't seem to notice a talk page message. Be a little patient. Anomie⚔ 12:26, 30 October 2026 (UTC)Reply

I have sent an email. Best, HouseBlaster (talk • edits • he/they) 14:09, 30 October 2026 (UTC)Reply

I've added {{nobots}} to the page. According to the brfa, the bot uses pywikipedia, so hopefully it respects that. --Chris 03:02, 31 October 2026 (UTC)Reply

I'll delete the task. -- Rick Block (talk) 23:04, 31 October 2026 (UTC)Reply

NervousBot (talk · contribs) (BRFA)

BOTREQ and VPP cited at the BRFA say "Bot that informs you when someone translates your article" and "Would anyone be opposed to a bot that notifies you when someone translates an article you made?", respectively. However, I was just notified about a disambiguation page that was certainly not translated. So, absolutely useless "somewhere, someone did something to some extent related to something you did 20 years ago". That's not what was originally requested at all. And why the hell is it spamming without an opt in? Max Semenik (talk) 10:44, 6 November 2026 (UTC)Reply

According to the BRFA, it's opt-in after one message. So unless you choose to opt in, that would be your last notification. – SD0001 (talk) 10:59, 6 November 2026 (UTC)Reply

Balance3641, out of curiosity, would it be possible to limit the timeframe in which the bot checks these cross-wiki creations? Not sure if Max is literally referring to a 20-year-old creation, but I do agree that if such a thing were to happen it's probably a wasted venture. If possible, maybe limit notifications to pages created in the last 3-5 years? (this isn't a BAG-mandated requirement, just a thought) Primefac (talk) 11:24, 6 November 2026 (UTC)Reply

I can look into doing this if there is any thoughts on what the filter should be, in reference to it being a disambiguation page, I do filter out pages tagged as disambiguation, but if they're not like Dmitriyev, it will slip through. I guess something like checking the last edit by that author on that page, could work in theory in some range? Balance3641 🍃 (talk · contribs) 11:36, 6 November 2026 (UTC)Reply

Dmitriyev is appropriately tagged as a set index article which probably should be exempted like disambiguation pages. Trialpears (talk) 12:22, 6 November 2026 (UTC)Reply

Yeah makes sense, will look at filtering those out. Balance3641 🍃 (talk · contribs) 12:23, 6 November 2026 (UTC)Reply

Any page transcluding {{dmbox}} should probably be ignored. Also probably any page containing a template that contains the EXPECTSHORTPAGE magic word. – Jonesey95 (talk) 16:31, 6 November 2026 (UTC)Reply

Thanks, that's the right fix and I've implemented it. Appreciate the input from you and Trialpears'. Balance3641 🍃 (talk · contribs) 02:34, 12 November 2026 (UTC)Reply

I have now set up an enwiki specific filter for these kind of pages (using the expectshortpage pageprop), I was just using disambiguation as the classifier. I'm yet to do anything regarding old articles. Wanted to get some more thoughts on that first. Balance3641 🍃 (talk · contribs) 02:32, 12 November 2026 (UTC)Reply