User talk:CoderThomasB/Archives/2023/April

Administrators' newsletter – April 2023
News and updates for administrators from the past month (March 2023).

Administrator changes
 * Gnome-colors-list-add.svg Aoidh
 * Gnome-colors-list-remove.svg KillerChihuahua



CheckUser changes
 * Gnome-colors-list-remove.svg Ferret

Guideline and policy news
 * A community RfC is open to discuss whether reports primarily involving gender-related disputes or controversies should be referred to the Arbitration enforcement noticeboard.

Technical news
 * Some older web browsers will not be able to use JavaScript on Wikimedia wikis starting this week. This mainly affects users of Internet Explorer 11.
 * The rollback of Vector 2022 RfC has found no consensus to rollback to Vector legacy, but has found rough consensus to disable "limited width" mode by default.
 * A link to the user's page will now appear in the subtitle links shown on . This was voted #17 in the Community Wishlist Survey 2023.

Arbitration
 * The Armenia-Azerbaijan 3 case has been closed.
 * A case about World War II and the history of Jews in Poland has been opened, with the first evidence phase closing 6 April 2023.

Discuss this newsletter

Subscribe

Archive Sent by MediaWiki message delivery (talk) 17:10, 4 April 2023 (UTC)

When is it appropriate to use content initially generated by an LLM?
Thanks for your important work maintaining Wikipedia! And for checking [my edit](https://en.wikipedia.org/w/index.php?title=WebAssembly&diff=prev&oldid=1149510377&diffmode=source) so quickly, and [letting me know that it might not be appropriate](https://en.wikipedia.org/w/index.php?title=User_talk:Chadoh#An_edit_that_you_recently_made_seemed_to_be_generated_using_a_large_language_model). I was unaware of the draft policy around LLM use; good to know.

I believe this is actually a case where using text originally sourced from an LLM is ok. I'll go through the guidelines and explain how my edit adhered to each, but let me start by explaining why I wanted this edit: I was working on documentation for a new blockchain project that uses Wasm, and I wanted a single word of that documentation to link to an authoritative website or page that lists all of the places where Wasm is used. One of the first places I looked was, of course, the Wikipedia page, specifically the Implementations section. I and my collaborators know enough about WebAssembly to know that this section of the Wikipedia article was far from complete! Failing to find a better list, I thought the best possible place for the information would be Wikipedia.

My teammate asked GPT-4 (I do not pay for it myself), and its list was a much better starting point than Wikipedia's. It had a couple small errors, which I corrected, and I added links to all of the projects it listed, so that any reader need only click them to verify that each does, in fact, use WebAssembly.

Ok, now let's go through those guidelines:

==== 1. You may ask neural networks to write original content and find sources. Even if such content was heavily edited, seek other alternatives that don't use the neural network's original content. ====

The "original content" was, in this case, mostly just a list. I edited this list and made it more appropriate for Wikipedia.

==== 2. You may use these neural networks as a writing advisor, i.e. asking for outlines, asking how to improve the paragraph, asking for criticism for the text, etc. However, you should be aware that the information it gives to you can be unreliable and flat out wrong. Use due diligence and common sense when choosing whether to incorporate the neural network's suggestion or not. ====

I used due diligence and common sense; I am an expert in the field.

==== 3. You may use these neural networks for copyediting and paraphrasing, but note that it may not properly detect grammatical errors or keep the key information intact. Use due diligence and do heavily edit the response from the neural network. ====

I did.

4. You are responsible for making sure that using neural network will not be disruptive to Wikipedia. Therefore, you must denote that a neural network was used in the edit summary.
I did.

==== 5. LLMs are not reliable sources. Unless their outputs were published by reliable outlets with rigorous oversight, they should not be cited in our articles. ====

I didn't cite one.

6. Wikipedia is not a testing ground for large language models. The use of Wikipedia for experiments or trials is forbidden.
I did not use Wikipedia as an experiment or trial.

7. You must use LLMs to write your comments.
I did not.

Chadoh (talk) Chadoh (talk) 14:46, 14 April 2023 (UTC)


 * Thank you for reaching out and taking the time to write a response! My decision to revert your edit came from a discussion in the tree house, which you can find a link to here: https://en.wikipedia.org/wiki/Wikipedia:Teahouse#Spotted_an_edit_that_adds_text_in_an_non-encyclopedic_tone_using_an_LLM._Opinions
 * In that discussion, the main focus was that the edit made a prediction in wikivoice "so-and-so is going to happen" rather than stating who made the prediction "This person predicts that so-and-so is going to happen". On that discussion, you can also find some other feedback to take into consideration when making edits to Wikipedia. If you want to, you can take that feedback and make a new edit that complies with Wikipedia tone of voice and editing guidelines. I would also advise that you read Core content policies if you haven't done so already, alongside that tree house discussion. - CoderThomasB (talk) 00:42, 15 April 2023 (UTC)