Wikipedia:Articles for deletion/LLM aided design

LLM aided design

[Hide this box] New to Articles for deletion (AfD)? Read these primers!

(Find sources: Google (books · news · scholar · free images · WP refs) · FENS · JSTOR · TWL)

Disputed draftification. WP:DRAFTOBJECT applies. If this is notable, it needs WP:TNT ~~because it cannot be divorced from its creation by AI.~~ Wholly inappropriately sourced with unreliable sources, fails WP:V, which is a key tenet of Wikipedia. Previoulsy sent to draft with the rationale While not conclusively AI-generated, the writing style, structure, and tone are consistent with LLM-assisted authorship. It likely had human curation or editing layered on top of content produced or scaffolded by a large language model. Further, the references are almost all deprecated sources. 🇵🇸‍🇺🇦 Fiddle^Timtrent Faddle^Talk to me 🇺🇦‍🇵🇸 08:51, 19 June 2025 (UTC)[reply]

@Timtrent author @Manvi jha13 came onto IRC Live Chat asking for assistance with this. They've repeated the article was not created with AI: they state they are pursuing a PHD in this topic so wrote the draft as an academic essay instead of an Wikipedia article. Have given guidance, and assuming good faith. qcne (talk) 09:16, 19 June 2025 (UTC)[reply]

@QcneThank you so much for your message.

@Timtrent, thank you very much for taking the time to review my draft and for providing your feedback — I sincerely appreciate your efforts.

It is rather intriguing to see the draft being marked as AI-generated again. I have stated in my talk page for the article and would like the opportunity to clarify again that no content of the given page has been generated by AI. The AI tools have been used for vocabulary suggestion, but in no case for text generation. I believe that given the academic use and exploration of the topic, along with the fact that I am a PhD student mostly engaged in academic writing, gives the article a similar tone, which I have tried to improve since your suggestions. Please do let me know if there are any additional areas/sections/perspectives you would suggest for me to improve on.

Additionally, I have noticied that you have reservations regarding the citations? I believe all the citations are academic publications. Please let me know if and how I can improve them.

Thank you,

Manvi Manvi jha13 (talk) 09:29, 19 June 2025 (UTC)[reply]

Manvi jha13, in reference to your claim on Talk:LLM aided design that "The use of AI was limited strictly to very occasional language refinement", could you please disclose in full detail the extent to which you used an LLM to generate the article, including the content, section headings, references, and formatting? Additionally, could you please disclose the name and versions of the AI tool(s) that you have been using to edit Wikipedia, as well as whether you are using those tools to author your comments in discussions like this one? — Newslinger talk 20:29, 19 June 2025 (UTC)[reply]

Hi @Newslinger

When I state that "the use of AI was limited strictly to very occasional language refinement," I am referring specifically to minor assistance such as suggesting synonyms or checking for spelling and grammatical errors (ChatGPT-4o). Importantly, no AI tools were used to draft or generate any content or contextual material.

Additionally, I want to clarify that AI was never used in drafting or contributing to any discussions or comments. I reaffirm that at no point was AI employed to generate new text or ideas, thereby eliminating any concern regarding hallucinations or the reliability of the content. Manvi jha13 (talk) 20:51, 19 June 2025 (UTC)[reply]

Manvi jha13, you made the edit Special:Diff/1296403283 to the article within the last hour. How did you generate the references and the citation code that you added into the article? — Newslinger talk 20:59, 19 June 2025 (UTC)[reply]

What do you mean generate references? They are the papers I have read, most of them are initailly made available on Arxiv and later published via conferences or journals. Why would it be difficult to find them?

As for citiation code, it is a rather starightforward format one can write it themselves, in any case to simplyfy my work, I wrote a small python script that takes bibtex format citaion and converts to wikipedia style. This helps reduce manual effort, and ensures consistency. I’ve made sure all included sources are verifiable and meet the reliability standards expected here. Manvi jha13 (talk) 21:04, 19 June 2025 (UTC)[reply]

@Newslinger, My apologies, I missed to ping you in my response, please refer to my reply above. Thank you in advance. Manvi jha13 (talk) 21:13, 19 June 2025 (UTC)[reply]

If the citation code were generated with a Python script, it's not clear why the code would use plaintext instead of normalized citation templates such as {{Cite journal}}, or why it would mix wikitext formatting with Markdown formatting (which is not used by Wikipedia).

This article exhibits too many characteristics of LLM-generated content to remain in article space. I am unconvinced that "The use of AI was limited strictly to very occasional language refinement" when the the very first revision (Special:Permalink/1294545580) already shows heavy signs of being LLM-generated, including the excessive use of lists and the idiosyncratic use of title case that are associated with AI chatbots. Draftify, and the draft should not be moved back into article space without going through the Articles for creation (AfC) process. — Newslinger talk 21:45, 19 June 2025 (UTC)[reply]

@Newslinger Manvi jha13 (talk) 21:47, 19 June 2025 (UTC)[reply]

@Newslinger

Thank you for your feedback. I don't understand why a python script would be limited to citation template, it would be able to take input and produce results based on how I program it. So I respectfully but completely disagree with this claim of yours.

Additionally, as I already stated, the use of ChatGPT was restricted to the use for checking grammar and spelling errors. To highlight the procedure goes like- I write a draft -> I pass it to ChatGPT with a prompt asking to fix any spelling or grammatical errors in the given text and just use that. This procedure in no way known to me generates new text. Additionally, in order to clarify again, this is the topic I am working on for PhD, the academic tone and style (including the usage of lists and detailed descriptions) is thus a result of the same Manvi jha13 (talk) 21:51, 19 June 2025 (UTC)[reply]

Do you have the version of your draft before you processed it with ChatGPT? — Newslinger talk 21:53, 19 June 2025 (UTC)[reply]

@Newslinger

Thank you for your question.

I would not have the article as a whole but yes I can get all the paragraphs I processed through the ChatGPT history. Would you like samples or screenshots (or other methods you deem satisfactory for proving, since that is what we are doing here)?

Honestly it is a bit intriguing to see how intolerant the Wikipedia community is of the academic community and their writing style. Manvi jha13 (talk) 21:59, 19 June 2025 (UTC)[reply]

If you could provide the pre-ChatGPT content in text form on the article talk page, Talk:LLM aided design, that would help establish that the article is not LLM-generated and also help editors improve the article by having your original writing available to reference.

The Wikipedia community appreciates the academic community in general, but many Wikipedians have a negative view of LLM-generated content. On Wikipedia, articles are expected to conform to the Manual of Style, and LLM-generated articles almost always deviate from the style guidelines in much more distinct ways than the average new editor would.

To clarify my previous comment, I did not say that a Python script would be limited to generating citation templates, although I do find it unusual that your script converts citations to "wikipedia style" by partially outputting Markdown instead of using a normalized citation template format. — Newslinger talk 22:12, 19 June 2025 (UTC)[reply]

@Newslinger

Thank you for your feedback.

Sure I can add pre-ChatGPT text for reference, just to clarify, do you expect the entire article or a few paragraphs would be enough?

Additionally for the python script, I do not use any libraries, my script simply takes the BibTex(easier to extract from), extracts details like paper name, author name etc.. and simply arranged them in a template I give. The template is the one I found to be the best fit for my scenario, it can be heavily varying from the general trend but I don't think that should be an issue? Manvi jha13 (talk) 22:40, 19 June 2025 (UTC)[reply]

If you are able to post the entire pre-ChatGPT article, that would be preferred as it would be most helpful to all interested editors. For your citation script, I highly recommend revising your script template to use Wikipedia's Citation Style 1 templates to ensure that it consistently meets Wikipedia's citation style guidelines. — Newslinger talk 22:51, 19 June 2025 (UTC)[reply]

Hi @Newslinger

I have added a sample in the talk section of the article. Please refer to it for context. I decided not to include the entire article, as I did not want to create a lengthy and potentially cluttered post there. However, if you still have any reservations about the use of AI in the article based on the example provided, please let me know.

Additionally, I found the article WP:CHATGPT, which clearly states that using AI to refine text is acceptable, as long as the content does not involve hallucinations, inaccuracies, or unverifiable claims. Given that the text in this article has been thoroughly reviewed and all sources are properly cited, I would like to ask if you have identified any instances where this might have been an issue? Manvi jha13 (talk) 02:04, 20 June 2025 (UTC)[reply]

Comment Thank you Qcne. I think that must be interpreted as Manvi jha13's opinion that it should be kept. This does not address the lack of WP:V in the nomination. I will accept their assurance about AI generation in good faith and strike that part of the nomination. It has now been drafified twice, which is one more time than DRAFTOBJECT allows. I do not feel it may be returned to draft space without a full consensus under these circumstaces, crcumstances whcih we would not be in without unilateral moves to mainspace (allowed, but unwise in this case). It may, however, be spared that via WP:HEY. 🇵🇸‍🇺🇦 Fiddle^Timtrent Faddle^Talk to me 🇺🇦‍🇵🇸 10:18, 19 June 2025 (UTC)[reply]
As nominator I have no objection to consensus based draftification, though I would prefer an assurance that, if sent back to draft, the creating editor will submit for review and work with the outcome of that review and any further iteration. That might be a closure condition, in an ideal world. [[If WP:HEY has happened pre closure then it shoul dbe retained. If I am notified I will consider withdrawal. 🇵🇸‍🇺🇦 Fiddle^Timtrent Faddle^Talk to me 🇺🇦‍🇵🇸 21:08, 19 June 2025 (UTC)[reply]
Note: This discussion has been included in the deletion sorting lists for the following topics: Language, Engineering, and Computing. Skynxnex (talk) 14:46, 19 June 2025 (UTC)[reply]
@Timtrent

Thank you very much for your thoughtful feedback and suggestions. I have revised the article accordingly. The updated version no longer includes arXiv or other non–peer-reviewed sources. I hope these changes help improve the article's quality and bring it closer to Wikipedia's standards for verifiability and reliability. Manvi jha13 (talk) 20:22, 19 June 2025 (UTC)[reply]
Draft: is the best option. Unfortunately, it's nearly entirely sourced to arXiv articles, which are not reliable sources. Pre-prints, meaning they've not been peer-reviewed yet. Once they get published, they would have to then show reliable sourcing. This article is also perhaps a bit too technical for a general audience. Needs a rewrite and better sourcing at this point. Oaktree b (talk) 15:23, 19 June 2025 (UTC)[reply]
or let it incubate offline and submit it for the AfC review. This wouldn't pass as is anyway. Oaktree b (talk) 15:24, 19 June 2025 (UTC)[reply]

Comment: Article should be improved, then in the longer term merged with AI-driven design automation. This is another new page, with a more general overview (not all AIs are LLMs). Both pages have issues, but the topic is surely worth keeping. LouScheffer (talk) 20:44, 19 June 2025 (UTC)[reply]
Hi @LouScheffer,

Thank you so much for your valuable review. I would greatly appreciate your guidance or suggestions on how the article could be improved.

While AI-driven design automation does involve hardware design, it is fundamentally different from LLM-aided design. AI-driven automation typically refers to techniques like MLIR or the use of Bayesian optimization and supervised/unsupervised/reinforcement learning to improve stages of the design process. However, its scope is generally limited to optimization rather than generation.

In contrast, LLM-aided design focuses on the ability to generate descriptions, code, and even complete designs from natural language input; something beyond the capabilities of traditional AI-driven automation. This distinction, I believe, is key to understanding the scope and novelty of LLM-aided approaches. Manvi jha13 (talk) 21:12, 19 June 2025 (UTC)[reply]

TNT Are sure this entire article is not LLM generated? It has a weird, unencyclopedic promotional tone. InvisibleUser909 (talk) 21:51, 19 June 2025 (UTC)[reply]

Comment (with no shade intended to User:Manvi jha13): I am interested in the assertion, "The AI tools have been used for vocabulary suggestion, but in no case for text generation." Vocabulary is part of text, and suggesting it entails generation, does it not? I am interested because part of my day job is to teach writing courses, and I often hear from students things like, "I didn't use AI. I only used <LLM-based app> to <do writing-related thing>." Again, with no shade to Manvi jha13, it seems to me that the definitions of terms such as AI, LLM, and generate are currently unsettled. This is something that might eventually be mentioned in this or a similar article (though, of course, only after it has been discussed in reliable secondary sources). Cnilep (talk) 01:21, 20 June 2025 (UTC)[reply]
Hi @Cnilep

Thank you so much for your feedback and interest in the topic. I'd like to offer some insights based on my understanding and research into LLMs so far.

To the best of my understanding, it would be considered "text generation" in the context of Wikipedia if the entire article or part of it were artificially created, which could potentially lead to false information or hallucinations (a known risk even with the latest LLMs). However, when the use of an LLM is solely for refinement purposes- such as improving grammar, suggesting synonyms, or rephrasing sentences- it's comparable to using a thesaurus tool or the inbuilt features in MS Word/Grammarly that flag grammatical issues and suggest more suitable word choices. In my view, this does not lead to the generation of entirely new or potentially inaccurate information.

Many people are opting for AI tools over MS Word or Grammarly because they can save a lot of time in the writing process. However, after reflecting on the depth of the discussion on this page, I'm starting to wonder if that time saved is worth it! Manvi jha13 (talk) 02:13, 20 June 2025 (UTC)[reply]