User talk:Garkeith

Please refrain from making edits generated using a large language model (an "AI chatbot" or other application using such technology) to Wikipedia pages, as LLM-generated content is prohibited on Wikipedia. Your edits have been reverted. If you would like to experiment, please use your sandbox. Thank you. glman (talk) 13:08, 28 July 2026 (UTC)

There is currently a discussion at Wikipedia:Administrators' noticeboard/Incidents regarding an issue with which you may have been involved. Magnolia677 (talk) 11:02, 29 July 2026 (UTC)

Please stop. If you continue to post comments on Wikipedia generated using a large language model (an "AI chatbot" or other application using such technology), as you did at Merchants National Bank Building (Clarksburg, West Virginia), you may be blocked from editing. Jako96 (talk) 22:04, 29 July 2026 (UTC)

You have been blocked indefinitely from editing for continuing to use LLM to create articles. See Wikipedia:Administrators' noticeboard/Incidents#User:Garkeith..

If you believe that there are good reasons for being unblocked, please review Wikipedia's guide to appealing blocks, then submit a request through the unblock wizard or through your talk page.  — rsjaffe 🗣️ 21:27, 31 July 2026 (UTC)

Could I respectfully request not to be blocked forever. I have been a long time editor literally like 20 years. I understood that I needed to stop using LLM and I was working on a new article which I used a template that I had created in the past with ChatGPT but if you actually look at the draft article I was working on everything was adding was done manually. I was not trying to slide that by, I was trying to get back to basics and do it correctly. Please have another look at that article Draft:John Simpson (trapper) I was working on if you are able and see I was building it correctly. I am not a trouble maker and I am not looking to cause problems. I enjoy working on pages related to my hometown and I got lazy admittedly. I want to get back to doing honest work please. Garkeith (talk) 15:40, 2 August 2026 (UTC)

Thank you for being clear as to what happened. You are unblocked with the condition (one that everyone here must follow) that you do not use an LLM to generate text. — rsjaffe 🗣️ 15:50, 2 August 2026 (UTC)

Understood, Thank you, I did not understand what a big deal it was until this rush of complaints about me. I am in full understanding now and I will adjust my behavior. Garkeith (talk) 16:10, 2 August 2026 (UTC)

Yes, the attitude towards LLM output has become very unforgiving here because of the amount of problems that it has caused. I appreciate your attitude towards all this and hope you continue to be a valued contributor to the encyclopedia. — rsjaffe 🗣️ 16:36, 2 August 2026 (UTC)

Can you explain Diff/1366466246, were you being truthful or not? ‑‑gurkubondinn 12:17, 3 August 2026 (UTC)

I still feel I was being truthful there, but like I have said previously, I was not aware of what a big deal this was. I am sure in the correction of spelling and grammar some of my sentences we edited by the AI. I now understand how touchy everyone here is about this subject and I am going to use AI in a very different way and follow the rules. I am not looking to draw this much attention, I enjoy editing not arguing with everyone. I am correcting my actions. Garkeith (talk) 13:44, 3 August 2026 (UTC)

Good, thanks. I just wasn't sure what you'd meant. I would just recommend using the spell checker in your browser, because it is really good (and unlike an AI chatbot, it will never alter you sentence or change the meaning of it). ‑‑gurkubondinn 14:14, 3 August 2026 (UTC)

An automated process has detected that when you recently edited Bridgeport, West Virginia, a link pointing to the disambiguation page Joseph Johnson was added.

(Opt-out instructions.) --DPL bot (talk) 18:04, 9 August 2026 (UTC)

fixed Garkeith (talk) 18:09, 9 August 2026 (UTC)

Hi Garkeith, I volunteer at the AI noticeboard and I checked your August 10 edit to Harrison County, West Virginia. The sourcing is clean, and the added material is a great contribution--well done!

We are wondering about cleanup of older material. There are articles with definite signs of LLM use that need to be cleaned up. If you would be willing to do that yourself, the best approach would be to move them to draft or user space so that you could work on them at your leisure. If you don't feel prepared to do that, we could really use some help establishing when LLM use started and ended, so we can focus on the potential problems and not disturb other articles.

My advice if you decide to clean up LLM articles is to cut them down to a stub, then build back up with human text only. They tend to have subtle falsehoods, misrepresentation of the sources, and peacock wording that are a lot of work to copy-edit out. (I've done it several times and come to the conclusion that stubifying is actually easier.)

If possible, could you (a) let us know if you can clean these articles up, and (b) let us know which articles, or at least time periods, you were using LLMs for? Or you could post to the messageboard thread at WP:AINB#Garkeith. Thank you! If I can be of any help please let me know. M kuhner (talk) 22:23, 10 August 2026 (UTC)

My current work has been the result of me working to fix some of my mistakes. I did not use it that much because I have not been that involved in editing for a while, but I did use it some. I really am not sure I could like make you a list. I had definitely used it on the Clarksburg page pretty heavily, that is why I dedicated myself to fixing that page. Then when I started getting warmed up I moved on to Bridgeport, and Shinnston WV. Then after I had done all that work it really made it easy for me to jump onto the Harrison County page because I already had so much of the work done. My Merchants Bank article that was deleted I know I was guilty of using it for most of that. I hope to remake that article and do it right this time. I am also needing to go back and work on the Robinson Grand Performing Arts Center, I am pretty sure it did most of that, or at least buffed up my work. I will need to research to find other pages I used it on. I am working on that and fixing my problems hopefully. I did not realize what a big deal it was until so many people were suddenly mad at me. I was just doing my thing. I actually trusted it way too much. Going back and reviewing my own work as I know people are seeing now with my new understanding of why it is a problem, I see how sloppy my work was. I will try to figure out what else I work on in that very small time period I did several pages around those two I mentioned. That is when I was using it. Garkeith (talk) 01:14, 11 August 2026 (UTC)

Thank you for the honest and cooperative response! (Seriously, that's not what we usually get.) See what you can do to identify the problem articles. I will hold off on making a cleanup tracker until we hear from you.

I also had the experience of thinking "well, this isn't so bad"--until I started working in cleanup and doing source-to-text analysis, and realized how many little falsehoods were creeping into LLM articles. It sounds so polished but it's all just surface with no truth underneath. I did a 100+ article case about science fiction novels, and as I checked article after an article I kept hearing "What is this plot summary? Did he read the same book I did? How can this be so messed up?" It was really an eye-opener.

If you would like someone to look at an article after you clean it up, I've done that for other editors, or someone else at AINB likely would. Drop a note if it would be helpful.

M kuhner (talk) 01:36, 11 August 2026 (UTC)

I think my intention at the time was to make all the landmark links on the Clarksburg, West Virginia page into articles instead of references without wikipedia pages. I tend to focus on my local history. So those are the ones I will be working on fixing. I do not think it was that much so I think I can handle repairing them. I really want my local pages to be good so I don't want them all being hunted down and deleted because I was on like a 2 week bender in love with ChatGPT. I will get it fixed. Garkeith (talk) 01:45, 11 August 2026 (UTC)

I think that's great. However, I have fixed LLM articles before and it is hard. I strongly suggest you get a second opinion on at least the first one you do. I am currently doing this with someone who writes about African activists, and it has taken 2-3 tries to clean each article up. (Really makes you appreciate that the "labor saving" is not saving anything at all....)

Best wishes, M kuhner (talk) 01:54, 11 August 2026 (UTC)

I already did a pretty through cleanup of Clarksburg, West Virginia, Bridgeport, West Virginia, and Shinnston, West Virginia. I don't specifically remember if I had been the one to do it all, I didn't check the changes to make sure it was my work, but a through check of those articles showed a lot. It was probably mostly me, In that same time period, especially in the history sections. They are pretty clean now. Feel free to check those out. I dont know what tools people like you use, but I taught ChatGPT all the rules of Wikipedia and fed it all the ai signs articles I could find and had it help me learn what to focus on fixing and what to look for. So I'm actually using it to fix it's own mess. Don't worry though, I'm not letting it do the writing! I am just using it to see what needs fixed. Garkeith (talk) 02:26, 11 August 2026 (UTC)

Some of the other editors at AINB use Pangram or other LLM detectors, but I just check sources against text by hand, plus look at AISIGNS. We also generally have at least two experienced editors look at each case.

I'm not confident ChatGPT can help you get the source/text matches right, or it would have done that with the text it was writing. LLMs are just really bad at that.

I'll have a look at the articles. M kuhner (talk) 02:32, 11 August 2026 (UTC)

Oh! A tool I do use, constantly, that might help you as well is mw:Who Wrote That? which is a browser plug-in that lets you easily see who wrote each word in an article. If you locate a line you wrote and click on it, all lines by you will light up. Super helpful. There are installation instructions at the link.

My process when there's a lot of small LLM edits in an article is to have one browser window with Who Wrote That? running and another window with the article open for editing. M kuhner (talk) 02:35, 11 August 2026 (UTC)

This is a very valuable tool, I have been playing with it this morning. Garkeith (talk) 14:54, 11 August 2026 (UTC)

with the specific instructions it finds a lot. It is kind of crazy. I will look at that tool. I think it makes a big difference between GPT 2023 vs 2026. It is significantly smarter, but you are right you can never fully trust it. I've learned that for sure. Garkeith (talk) 02:35, 11 August 2026 (UTC)

The two favorite tricks of 2026 LLM are very subtle and nasty. One is to say "Many people said" when actually it was only one person. Lots of variants but always the basic idea is that it exaggerates the number. The other is to take a statement that's really in source A, but claim it's in source B. Those are incredibly hard to catch because if you read both A and B, you remember reading this statement and don't realize it is attributed to the wrong source. But occasionally it becomes a very serious error, if B is being sourced for something they would not agree with.

Another problem is that (like some people) they hate to say they don't know. If you ask for a plot summary of a book that is not on the internet, they will make one up.

(I also have a funny story about this. I once asked Perplexity AI if it was possible that "my name" the chessplayer and "my name" the scientist could be the same person. It said no, of course not, and then got increasingly nasty when I disagreed with it, accusing me of bias! But in fact both of them are me, and a human could have figured that out.) M kuhner (talk) 02:46, 11 August 2026 (UTC)

I know where it learned that! "Many people are saying..." -Donald Trump Garkeith (talk) 02:50, 11 August 2026 (UTC)

Argh!

You're probably right. In general they seem to have been tutored by bullshitters. M kuhner (talk) 03:09, 11 August 2026 (UTC)

I checked the first two paragraphs in "Early settlement and development" which are sourced to (source 14).

There is a general problem that the text is very, very similar to the source--same ideas in the same order, many of the same words. I am an LLM hunter, not a copyvio hunter, but this seems dangerously close.

There are specific points where the text adds things the source doesn't mention. "Hunter and trapper" in the article is sourced to "trapper" in the source. (Probably he did hunt, but it doesn't say that.)

Article says "chartered by the Virginia General Assembly in 1785" but source only says "chartered in 1785."

That's about one issue per short paragraph. I'm afraid the whole thing has to be checked at this level, which will be quite time-consuming. ChatGPT cannot do it for you: if it could, it wouldn't have made these mistakes in the first place.... I could do a few more paragraphs tomorrow if that would be helpful, but every single claim will need to be checked. M kuhner (talk) 02:57, 11 August 2026 (UTC)

I knew about the hunter trapper thing but it seemed so obvious I went with it. I was wondering if I should change it because it didn't say it. So that is on me. Garkeith (talk) 11:49, 11 August 2026 (UTC)

I did another run through of the whole page. After seeing what you found that quickly I am trying to adapt to what I know you are looking for. I did another pass and cleaned up language on the whole page. A few other things I knew were questionable but I was hesitant to change are now corrected. It is a work in progress but it is WAAAAAY better than it ever was previously. Garkeith (talk) 14:53, 11 August 2026 (UTC)

An automated process has detected that when you recently edited Italian Heritage Festival, a link pointing to the disambiguation page Morra was added.

(Opt-out instructions.) --DPL bot (talk) 17:55, 16 August 2026 (UTC)