User:LWG
|
.svg){kind=small}
.svg){kind=small}
.jpg){kind=small}
|
.svg){kind=small}
|
My primary areas of off-wiki expertise are descriptive linguistics and mountaineering. On-wiki, I primarily contribute via source verification and cleanup of old articles with stagnant POV disputes. I also watch articles on a variety of subjects that happen to interest me or that I have edited in the past, and have recently gotten more involved in cleaning up AI-hallucinated material.
I prefer to focus on wiki-space improvements to articles, so if I fail to reply to you in a discussion it's probably nothing personal, just that my limited capacity for talk-space discussion has been exceeded for the month.
On your right you can find some userboxes with basic information relevant to my wikipedia activities. If you want to know more about me, I invite you to look at my edit history and form your opinions from that. While you are at it, feel free to correct any mistakes I may have made!
Quick link to query likely to contain AI-generated content in need of scrutiny
Ideas that might be the seed of future essays:
- The use of Wikipedia for training AI models makes the rapid removal of bad content more important than ever.
- The tendency for inexperienced users to fill content holes with AI-generated slop makes the production of high-quality human-written content to fill those holes more important than ever.
- All proposed policies have two layers of consideration: what do we want Wikipedia to be like, and what do we think other editors will do when they read this policy?
- Bold editing and collaboration isn't the same as offloading work to others.
- Debates between stating things in Wikivoice versus attributing claims are almost always a waste of time - resistance to attribution is almost always some form of WP:RGW, and attributing claims almost never harms our readers.
- Reticence about things like AI-assisted editing/grammar checking/automated formatting tools is never constructive and always violates WP:AGF, as the only reason to hide info about your workflow from another editor is that you assume the other editor will do bad faith things with that information.
- The longer an editor stays on the Wiki, the more they understand PAGs, but also the more time they spend talking about/making new PAGs and essays and the less time they spend writing constructive content.
- Assume Good Faith does not mean Assume Competence. We can assume good intentions while doubting the quality of content, especially if there is a pattern of past bad content from the same source.
- Assume good faith is the baseline, if we really want to get stuff done we should also demonstrate good faith and invite boldness.
Essay I wrote:
10 Wikipedia Policies, Guidelines, and Expectations That Your ChatBot Use Probably Violates
Happy editing!
Old AI thoughts
|
|---|
|
Inclusionism in the age of AI Over my many years of involvement on Wikipedia, I've generally taken the view that retaining content is better than removing it, even if it is poorly sourced, because Wikipedia is a long term project, and content that can be improved should be given the chance to improve. However, this philosophy needs to be tempered somewhat now that the internet is full of AI-generated content. Wikipedia is a key source of training data and input for AI tools, and as a result false information on Wikipedia will rapidly be incorporated into AI's knowledge base, mingled with more reliable information, and then repeated as true. This accelerates and exacerbates the process of Citogenesis, where unsourced and incorrect information originating on Wikipedia is repeated in other venues and eventually makes its way into the sources wiki editors use, thus locking false information in place in a circular citation loop that can be very hard to get rid of. In light of this, I encourage wiki editors to be more aggressive in removing dubious and unsourced content. As an alternative to total loss of content in cases where improvement may be possible, I recommend moving content to talk or user space while it is being worked on. *2025 update* Unfortunately, the circular poisoning of AI trained on Wikipedia being fed back into Wikipedia has now dramatically accelerated due to the insertion of content generated by LLMs directly into Wikipedia articles. When I wrote the paragraphs above my primary concern was that sources who use AI to generate content would be cited unknowingly by Wikipedia editors. Now, the situation is that content is being generated by editors who don't even read sources at all, but simply ask their LLM to generate both the content and the citations. This is unacceptable. Please do not insert any content into Wikipedia that is not based on sources that you as the editor have personally read. It is inherently impossible to satisfy WP:V if you have not read the source you are citing. By inserting AI-generated text into Wikipedia, you are not only doing a disservice to readers of Wikipedia, you are also contributing to the poisoning of future AI training datasets. |
Old project: processing POV disputes
|
|---|
|
POV Processing One of my major contributions to the wiki has been going through backlogged POV dispute tags. The vast majority of these tags were either inappropriately added or the dispute in question ended a long time ago. The first step to focusing attention on the remaining serious POV issues on the wiki is the elimination of all the spurious or dead tags. This has been done for all the dated subcategories through the end of 2023, meaning that most/all of the remaining articles from 2022 and before may have outstanding POV issues that require more in-depth editing/research to resolve. That is a great place to focus improvement efforts if you would like to help write/edit content! If you want to help me process the remaining bulk of unassessed tags from 2023-2025, there's a box at the top right of this user page with the backlog sorted by date. Some general guidelines:
See this page for more guidance on tag removal. If you feel that one of my tag removals was inappropriate, feel free to replace the tag. If you do, please also contribute to the article talk page explaining what still needs to be done before the tag may be removed. Otherwise expect a talk page message from me requesting clarification. Why do this? If someone is complaining about the POV of an article, that means that article is important to someone. If things are important to someone, they are probably interesting. In the course of working through the POV backlog I have discovered all sorts of fascinating new bits of knowledge about obscure controversies and local geopolitical conflicts. This is fun! Articles that are important to someone also do more harm if they are of poor quality, so they are a good place to focus cleanup efforts. Unfortunately, such a large percentage of the outstanding POV tags have either already been resolved or were unexplained drive-by tags from the beginning that the important and interesting articles are buried in the noise. My hope is that if the backlog of noise can finally be cleared, future POV disputes can receive the attention they need to be promptly and appropriately addressed, which would be of great benefit to Wikipedia as an encyclopedia and to the Wikipedia community in general. Progress so far When I first took interest in this endeavor, there were tagged articles going back all the way to 2006. I have processed thousands of articles over the years, but thousands more have been tagged, so that as of today there are still about 2119 articles in the backlog. Still, the fact that my extremely part-time efforts have more or less single-handedly prevented the backlog from increasing over the past 10 years suggests that if the backlog could ever be eliminated, the handling of incoming POV disputes would be a very manageable task for a handful of dedicated editors. Recently other editors have joined in the work and the reduction of the backlog is accelerating such that totally clearing the backlog of stagnant or resolved disputes might be feasible in the near future. My long-term vision would be to see the backlog cleared of inactive tags and the formation of a sort of Dispute Response Team consisting of editors who are experienced in mediating content disputes and are able to engage in disputes as they arise to guide the involved parties towards constructive consensus building and away from contentious edit warring. This would improve the quality of the wiki content as well as help with editor retention as fewer new editors would be immediately pushed away as soon as they step into a content dispute. Please help! |