User talk:Daniel G Rego

You have been blocked indefinitely from editing because your recent edits appear to be disruptive. Please note that this is not necessarily an assessment of faith or character, and we welcome good faith edits that are also constructive and in line with our policies.

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. Star Mississippi 13:26, 31 May 2026 (UTC)

This user's unblock request has been reviewed by an administrator, who accepted the request.

Daniel G Rego (block log • active blocks • global blocks • contribs • deleted contribs • filter log • creation log • change block settings • unblock • checkuser (log) • SI)

Request reason:

The following request was written through the unblock wizard.

Can you explain, in your own words, what you were blocked for?

I was blocked due to improperly following the policies of WP:LLM/WP:NOLLM.

I am a native speaker of both English and Portuguese.

I noticed a range of articles on subjects with which I was familiar that fell into one of the following categories:

a) The subject had an article in Portuguese, but no corresponding article on English Wikipedia; or

b) The English Wikipedia article existed, but was a stub, was poorly referenced, was poorly written, and/or contained factual inaccuracies. In many cases, the articles already had maintenance notices identifying these issues.

I proactively took action to address these problems through what I believed to be the following proper five-step process:

1. I translated the corresponding, more complete, and better-developed content from the Portuguese articles into English using an LLM, in accordance with what I understood to be permitted under the WP:LLM policy:

“Editors are permitted to use LLMs to translate articles from another language’s Wikipedia into the English Wikipedia, but must follow the guidance laid out at Wikipedia:LLM-assisted translation.”

2. I carefully reviewed each translation to ensure that there were no hallucinations, abstractions, additions, or unsupported changes from the original Portuguese article to the English version. 3. I manually adapted the translated text from a mechanical translation into proper English. Some constructions, idioms, figures of speech, and cultural references had been translated too literally and did not read naturally for an English-speaking audience or for readers unfamiliar with the relevant cultural context. 4. Because the translation and adaptation work had been done in Microsoft Word rather than in draft space, a sandbox, or another wiki environment, the text did not yet have full wiki formatting. To expedite formatting, references, and related cleanup, I used the following exact prompt to format the material for Wikipedia:

“I need a final pass, line by line of my revision to this article against the latest English Wikipedia guidance, including WP:LAYOUT, WP:CITE, MOS:REFPUNCT, MOS:DATEUNIFY, WP:DATERANGE/MOS:DATERANGE, WP:NBSP/MOS:NBSP, WP:NDASH/MOS:NUM, citation style consistency, citation order, named-reference syntax, correct placement of explanatory notes, duplicated references, missing or undefined references, unused references, malformed URLs, template syntax, table syntax, image syntax, list formatting, section order, categories, DEFAULTSORT, hatnotes, portals, navboxes, and any other syntax, typo, formatting, or layout issues you find. Do not omit sections, images, references, lists, tables, or categories unless there is a clear English Wikipedia policy reason to do so, and explain any such omission.”

I believed that this was permissible under the current WP:LLM policy, which states:

“Editors are permitted to use LLMs to suggest basic copyedits to their own writing, and to incorporate some of them after human review, provided the LLM does not introduce content of its own. Caution is required, because LLMs can go beyond what is asked of them and can change the meaning of the text such that it is not supported by the sources cited. Examples of basic copyedits include spelling, punctuation, and capitalization.”

5. I then published the revised English Wikipedia articles, or created the new articles, with proper full edit summaries and talk page notices, including attribution for the Portuguese Wikipedia translation.

All of my proposed new articles were reviewed by New Page Patrol and accepted without complaints or adverse comments.

Several of them were independently rated as “B”-class articles. In some cases, I received compliments from other editors. A few active WikiProject Brazil editors also made subsequent edits to some of the articles.

Later, another editor raised the issue at the AI noticeboard because the prompt beginning with “WP:LAYOUT, WP:CITE, MOS:REFPUNCT, MOS:DATEUNIFY, WP:DATERANGE/MOS:DATERANGE, WP:NBSP/MOS:NBSP, WP:NDASH/MOS:NUM, citation style consistency” had accidentally been published along with the article text.

I tried to explain that the articles had been translated with full Portuguese Wikipedia attribution in both the edit summaries and the talk pages, and that the prompt had been intended strictly to catch technical, style, syntax, formatting, and related cleanup issues. The article content, sourcing, organization, and final wording remained my full responsibility. My revisions substantially improved the articles from the state in which they had previously existed.

Within less than 24 hours of my publishing that explanation, editors at the AI noticeboard proceeded to revert all of my edits, including edits where I had simply corrected egregious wording manually. My account was indefinitely blocked by an admin. They also proceeded to delete all of the articles I had created, even though those articles had already been reviewed and, to the best of my understanding, contained no AI-generated content beyond what I believed was permitted under WP:LLM and WP:NOLLM, including machine-assisted translation and machine-assisted copyediting.

If unblocked, what edits would you make and what (if applicable) would you do differently?

If unblocked, I would proceed much more cautiously and transparently.

First, I would not immediately restore the reverted material or recreate the deleted articles. I would treat the reversions and deletions as a serious signal that other editors had concerns about my process, even where I disagreed with the outcome. I would first seek consensus, ask for specific guidance, and work through draft space or article talk pages before making any substantial changes again.

For any future translation work, I would use draft space or my sandbox rather than preparing the text externally and then publishing it all at once. I would make smaller, more reviewable edits, with clearer edit summaries explaining exactly what I did. Where material was translated from Portuguese Wikipedia, I would continue to provide the required attribution in the edit summary and on the talk page, but I would also be more explicit about the workflow used. I would also be more careful to prevent any drafting notes from being accidentally published. Before saving any edit, I would review the diff carefully and make sure the article contains only article text, references, templates, and other appropriate wiki markup.

Substantively, I would focus first on low-risk cleanup edits to existing articles: fixing grammar, correcting obvious typos, repairing citation syntax, improving attribution, removing unsupported wording, and addressing maintenance tags. I would avoid creating new articles or making major expansions until I had rebuilt trust and received clearer community guidance on the acceptable process.

The main thing I would do differently is slow the process down. My intention was to improve weak, underdeveloped, or missing coverage of Brazilian and Lusophone subjects, but I now understand that the process also needs to be visibly reviewable by other editors. Going forward, I would prioritize transparency, smaller edits, source-by-source verification, draft review, and consensus-building over speed or volume.

Daniel G. Rego (talk) 21:14, 11 June 2026 (UTC)

Accept reason:

Per below. voorts (talk/contributions) 00:43, 15 June 2026 (UTC)

Why didn't you explain your process more clearly at the AINB thread? Why did you instead give vague/misleading answers? voorts (talk/contributions) 22:44, 11 June 2026 (UTC)

(Non-administrator comment) I'd also like to please ask whether you used it to write your appeal? In solidarity, Blue-Sonnet (I'm listening) 23:01, 11 June 2026 (UTC)

@Voorts:

You are right, in retrospect, I should have written my full rationale for the sake of transparency and clarity, instead of the shortened version I wrote in the AINB thread.

As a matter of fact, I had no idea of the institutional role or practical significance of that noticeboard, given the little time I have here is used to actually improve articles, so I never spend time beyond mainspace. (Actually, I don't spend any time on internet boards, forums, comment sections, etc).

Now, I do understand the importance of accountability and scrutiny. Since creating my account in 2008, I have edited under my real name and left my real life contact info, because I have always viewed that as part of transparency, the same approach I take in real life as an executive, academic, and board member.

Therefore, since I treat this as academia, even after politely asking for WP:CIV, I noticed the thread continued veering towards this sort of internet-culture-like combative, disrespectful approach, including name calling, and snap conclusions, instead of being concerned about discussing policy, analyzing facts, or engaging in a collegial, scientific-like discussion. That made me question the institutional role and purpose of the noticeboard. A day or so later, when I logged in again, my account was already blocked.

To answer your specific question directly: I did not explain my process more clearly at AINB because, at the time, I did not understand the noticeboard’s institutional role, procedural significance, or the consequences of an incomplete explanation there. I treated the thread as a hostile and informal internet discussion rather than as a quasi-administrative review requiring a full procedural account. That was a mistake. My answers were not intended to be vague or misleading, but they were incomplete because I was responding defensively to what I perceived as snap accusations rather than methodically setting out the entire workflow. In retrospect, I should have explained, from the beginning, the full sequence.

@Blue-Sonnet: If your comment is an inquiry about substance or style, please note that I write and act formally. To answer your specific question: no, since there was neither formatting nor translation involved. Thanks.

For clarity, this is the response I posted at AINB. In retrospect, I should have made it more descriptive from the outset, as I did it in our thread here.

I want to clarify my process here:

My understanding of WP:LLM/WP:NOLLM is:

"Editors are permitted to use LLMs to suggest basic copyedits to their own writing, and to incorporate some of :them after human review, provided the LLM does not introduce content of its own. Caution is required, because LLMs can go beyond what is asked of them and can change the meaning of the text such that it is not supported by the sources cited. Examples of basic copyedits include spelling, punctuation, and capitalization."

Therefore, after the work is performed, using assistance such as:

I need a final pass, line by line of my revision to this article against the latest English Wikipedia guidance, including WP:LAYOUT, WP:CITE, MOS:REFPUNCT, MOS:DATEUNIFY, WP:DATERANGE/MOS:DATERANGE, WP:NBSP/MOS:NBSP, WP:NDASH/MOS:NUM, citation style consistency, citation order, named-reference syntax, correct placement of explanatory notes, duplicated references, missing or undefined references, unused references, malformed URLs, template syntax, table syntax, image syntax, list formatting, section order, categories, DEFAULTSORT, hatnotes, portals, navboxes, and any other syntax, typo, formatting, or layout issues you find. Do not omit sections, images, references, lists, tables, or categories unless there is a clear English Wikipedia policy reason to do so, and explain any such omission.

In order to catch technical and style issues and similar cleanup issues. Therefore, the article’s content, sourcing, organization, and final wording are my full responsibility and I believe the revision improved the article from the state in which I found it.

But I understand why patrollers are being careful here. That's actually why I love reading Wikipedia, and (when I can, volunteer my time and money), because the quality here tends to be quite good.

And I get, I've been here for a long time, and I've seen my fair share of vandalism, low-quality editing, and all sorts of disruptions your team has to deal with it.

So, again I appreciate the concerns, but that being said, I rest my case.

Daniel G. Rego (talk) 00:19, 12 June 2026 (UTC)

After reviewing your edit history and reply, I still have some concerns. Can you please also answer the following questions?

You wrote: I carefully reviewed each translation to ensure that there were no hallucinations, abstractions, additions, or unsupported changes from the original Portuguese article to the English version.
- You've copied the response from the AI into at least three articles on separate occasions

The first diff shows the AI's reply right at the top, so this is the first thing visible when you look at the text. How did you miss the AI's reply if you've been carefully reviewing all your edits, considering it's the very first thing at the top of the page and the first thing that is visible? I don't understand how this could have happened on multiple separate occasions if you were carefully reviewing the AI's output before publication, so I would greatly appreciate some clarification on this point.

Or is it the case that you weren't reviewing the AI's output because you asked the AI to do it for you? Your appeal seems to indicate that the "final pass" was done by prompting Copilot to do it for you, but that would contradict your statement above that you did it personally, so I'm a little confused over what caused the AI's text to be included so many times.

I'm not asking to be difficult; if we don't understand why this happened, we can't prevent it from happening again.

Your appeal states Some constructions, idioms, figures of speech, and cultural references had been translated too literally and did not read naturally for an English-speaking audience or for readers unfamiliar with the relevant cultural context.
- All I can find on the noticeboards are editors voicing concerns that your edits include language and phrasing that's frequently used by AI, but nobody said the text itself was unclear or confusing. Can you please clarify who said this was an issue?

In solidarity, Blue-Sonnet (I'm listening) 01:04, 12 June 2026 (UTC)

Certainly, so the problem occurred between steps 4 and 5 of my "workflow:

1 - LLM Translation;

2 - "I carefully reviewed each translation to ensure that there were no hallucinations, abstractions, additions, or unsupported changes from the original Portuguese article to the English version."

3 - Manual adaption from mechanical translation to natural English prose;

4 - AI comes back on: "I need a final pass, line by line of my revision to this article against the latest English Wikipedia guidance, including WP:LAYOUT, WP:CITE, MOS:REFPUNCT, MOS:DATEUNIFY, WP:DATERANGE/MOS:DATERANGE, WP:NBSP/MOS:NBSP, WP:NDASH/MOS:NUM..."

5- Publication.

From step 4 to step 5 there should have been something in between. I copied the output from the final technical pass where the Word version became wiki source code without performing the additional global manual review that I should have performed after that step. That is how prompt response was accidentally included. I should have at least hit review before publishing, so it was a copy-and-paste and final-review failure after the technical cleanup stage.

I think the safeguard should have been a final manual inspection of the entire article text in the Wikipedia editor. Beyond that, however, one important clarification that I had pondered, but had not found an authoritative answer to, was whether using the following prompt:

violates Section 1 of WP:LLM, which states:

If the use of that prompt falls outside what the policy permits, then I unintentionally violated the policy in good faith based on my interpretation of it. Again, my apologies. Since I don't spend time beyond mainspace, in retrospect I should have probably checked any current "case law" that may been left on the discussion forums. Daniel G. Rego (talk) 02:10, 12 June 2026 (UTC)

(Non-administrator comment) Daniel, with respect, since you're a native speaker of both languages would it not be faster and easier to use a workflow like this:

Assess sources from Portuguese Wiki for compliance with English Wiki policies
Optionally, ask an LLM to write a bullet-point layout of the draft
Summarise the suitable sources into a draft

There may be stylistic and grammar mistakes, but it's so much easier for other editors to tidy those up (and many editors love doing that!) than it is to fix AI/LLM prose (which has to be completely rewritten). You already have to read the sources to make sure the LLM is summarising them correctly, then you have to rewrite everything it's written, and then you have to do yet another review to make sure it hasn't messed anything up after the 'style check'. You're putting in so much extra effort when I genuinely believe you could do a far better job than the LLM, and in significantly less time. Other editors would be happy to do the 'style check' (and already know the policies and guidelines, where LLMs like to make them up) if that's not one of your strengths. Drafts don't have to be perfect, they just have to summarise suitable sources accurately and cite them appropriately. Meadowlark (talk) 06:10, 12 June 2026 (UTC)

@Meadowlark: Hello, sorry for the delay. I’m not sure it would necessarily have been faster, but it clearly would have been better and safer. My rationale was to use what I believed was permissible under the guideline to expedite the process, since there are countless Lusophone-related articles in poor condition and WikiProject Brazil appears to be quite inactive. Obviously, though, given the controversy and the fact that all of the changes were reverted, including changes to articles that had absolutely nothing to do with that workflow, it ended up being a waste of volunteer time. I should not have used that approach. Daniel G. Rego (talk) 17:22, 14 June 2026 (UTC)

LLMs are so good at sounding confident and acting like they're helping, even when they're not. It's a delightful change to see someone who recognises that human writing is better for Wikipedia rather than insisting that LLMs are superior, and I applaud you for being willing to listen. It looks like you're about to be unblocked, so I wish you happy editing and hope to see you around! Meadowlark (talk) 06:25, 15 June 2026 (UTC)

Glad to see the unblock too, the trick is untangling everything and getting everyone on the same page! Hopefully this isn't too off-topic, but I've recently seen some research showing that the more confident an AI sounds about something, the higher the likelihood that it's completely wrong. In solidarity, Blue-Sonnet (I'm listening) 07:41, 15 June 2026 (UTC)

@Star Mississippi: any objections to unblocking? voorts (talk/contributions) 20:23, 14 June 2026 (UTC)

@Voorts I think @Daniel G Rego has shown a good understanding of why the community feels his edits/process are problematic, even though he didn't realize it at the time. No objection to a second chance. Thanks for the ping. Star Mississippi 00:42, 15 June 2026 (UTC)

User talk:Daniel G Rego

A cup of coffee for you!

May 2025

May 2026

Related Articles