DZ
DefZone.Net

Kien thuc cong nghe & giai tri

Wiki Article

Talk:ChatGPT

Nguồn dữ liệu từ Wikipedia, hiển thị bởi DefZone.Net

Xem trên WikipediaBài ngẫu nhiên khác

Current model status sourcing

[edit]

@Alenoach I see the model versioning status in the table is back. It is not clear to me what the source is for these values. Can you please clarify? I am considering making this my next GAN so if there is completely uncited content I plan to remove it. Czarking0 (talk) 04:01, 22 December 2025 (UTC)[reply]

The source is at the top of the "Status" column. I just updated it (the previous one was from when GPT-5 was released). Alenoach (talk) 23:24, 22 December 2025 (UTC)[reply]
This source does not mention 3.5 so I don't believe it can be considered the source for saying GPT3.5 is discontinued. Czarking0 (talk) 17:51, 26 December 2025 (UTC)[reply]
The source is exhaustive about the legacy models currently supported; the models that are not mentioned there are not available anymore on ChatGPT. It's not so clear to me why you would like to remove this column, is it just sourcing or is there another reason? Alenoach (talk) 19:01, 26 December 2025 (UTC)[reply]
Policy indicates that all claims need sources. The status of each model version is a claim and each needs a source or should be tagged with cn. If after a month or so of cn tag then I will remove claims on this page. Czarking0 (talk) 21:13, 26 December 2025 (UTC)[reply]
What I argue is that it is actually sourced: the source lists exhaustively the models currently available in ChatGPT, and indicates which of them are legacy models. If a model isn't available anymore, it has been discontinued.
Having a single source for the whole column like this is way better than a ref on each row of the column. And it seems to me that this column is really useful for being able to see at a quick glance which models are outdated or discontinued, which is important information that is almost entirely absent from the rest of the article. Otherwise, readers may for examples read the subsections on GPT-4 or o1 without realizing that they do not exist anymore in ChatGPT. I would suggest to precise though that it is only the status "in ChatGPT", as these other models still exist on OpenAI's API.
If you still want to remove it, maybe we could ping two or three major contributors to this article to ask if they have an opinion on this? Alenoach (talk) 22:09, 26 December 2025 (UTC)[reply]
Eh I'll just leave this discussion up. I see your point, if the GA reviewer has a problem with it we can address it then. Czarking0 (talk) 23:30, 26 December 2025 (UTC)[reply]

Potential good article nomination

[edit]

@Czarking0, you suggested at some point applying to WP:GAN for the article become a good article. Do you think the article is ready? The main issue I see is that the subsection "Law" is a little too long, maybe some of it could be moved to a law-related article. Alenoach (talk) 19:42, 26 December 2025 (UTC)[reply]

No, I think the computer science section is outdated. I also support your idea though I think AI in law would make a better article topic than strictly ChatGPT in law. FYI I also started Draft:Context Window while working on this. Czarking0 (talk) 21:10, 26 December 2025 (UTC)[reply]
I don't think I will have time to help on this, but it seems to be a good idea. I had multiple times the impression that an article on the concept of context window was missing. There is an existing French article on this which could potentially be used for inspiration, but which really lacks sources. Alenoach (talk) 22:24, 26 December 2025 (UTC)[reply]
And yes, it could be reasonable to add an article named "Artificial intelligence in law", but I don't have the expertise and time to create such an article. I considered potentially moving some of the content to the article Legal_informatics#Artificial_intelligence or Applications_of_artificial_intelligence#Law_and_government, but I don't know if it's a good idea. Alenoach (talk) 22:38, 26 December 2025 (UTC)[reply]
There I took a crack at it. What do you think? Czarking0 (talk) 04:42, 4 January 2026 (UTC)[reply]
It looks good. I made some refinements. Alenoach (talk) 12:23, 4 January 2026 (UTC)[reply]
I like it. Shall I start a joint nomination? Czarking0 (talk) 19:17, 4 January 2026 (UTC)[reply]
Yes, feel free to do it if you think the article is ready. Alenoach (talk) 07:19, 5 January 2026 (UTC)[reply]
Do you think the image "Three-stage large language model training workflow.svg" in the training section is good? I was thinking that maybe the font is a little too small for comfortable reading on mobile. We can ask Kjerish for a slightly bigger font if necessary. Alenoach (talk) 17:20, 6 January 2026 (UTC)[reply]
I don't have a strong opinion here. I recognize your concern. I think it is hard to make a good graph like this for mobile. That does not make it not worth having. Czarking0 (talk) 01:42, 13 January 2026 (UTC)[reply]
Not really considering a removal, rather whether the font should be made slightly bigger or whether it's already readable (the user who created the image was pretty receptive to feedback last time I asked, but I don't want to bother if I'm unsure it's an improvement for readers). Alenoach (talk) 22:35, 13 January 2026 (UTC)[reply]

ChatGPT Health

[edit]

Would this article be the best space to add a section about ChatGPT Health? OpenAI just announced this new add-on and I'm not sure where it should live or if it shouldn't. 🥑GUACPOCALYPSE🥑 17:49, 8 January 2026 (UTC)[reply]

Regional responses addition reverted

[edit]

Hi @Czarking0, I would like to better understand what you mean by '"aimed to influence" indicates this is not due weight for the main article' in the recent revert on my edit here. My addition is preceded by 3 other similar examples of 2024, by 3 different countries (China, Israel and Russia), so if you can further elucidate why you think it is undue and how would you write it otherwise. Thanks Josep a11 (talk) 21:43, 12 January 2026 (UTC)[reply]

My read of this section:
  • Actual impact of ChatGPT in China (much discussed in previous Talk page threads).
  • Actual ban and repeal regarding ChatGPT in Italy.
  • OpenAI actually banning state sponsored usage.
  • Actual FTC investigation and sanctions.
My understanding your proposed addition:
  • Israel attempts to influence ChatGPT
To me your claim may well be notable on another page but for this page it lacks the actual impact on the article subject. Czarking0 (talk) 01:40, 13 January 2026 (UTC)[reply]
When reading the section holistically, I agree with your summary, this is also what I read. However, I do not agree that the claim is not notable enough, nor that it lacks impact on the article subject. I placed the claim mainly to highlight that the risk that came with the advent of the RAG technique is also leveraged by state actors for influence. The new attack surface at play here, which involves tapping into knowledge databases to produce the desired output response by the model is explained by Wei Zou et al. (2024) https://arxiv.org/pdf/2402.07867, and it is not mentioned in the article whatsoever. So, I propose we include a new paragraph under the section with both links to the aforementioned study and the use-case that Israel did on 2024. Alternatively, another sub-section under "limitations" might be considered. Josep a11 (talk) 11:11, 13 January 2026 (UTC)[reply]
It does not seem related to ChatGPT in particular, rather about chatbots in general. Alenoach (talk) 23:44, 13 January 2026 (UTC)[reply]
I guess you refer to generative AI bots. If your point is that if it not ChatGPT-only, then it should not be placed in the present page, then there are a lot of existing paragraphs from the article that can be deleted, as are even more vague than the present, such as the one citing the American Bar Association that generally talks about generative AI. The paper above (and other sources not cited here) does note that it affects ChatGPT, and they were explicitly tested on ChatGPT, so I disagree with your point; I vote that it should be appended in some shape or form. Josep a11 (talk) 11:16, 14 January 2026 (UTC)[reply]
I agree that the ABA paragraph is not specifically about ChatGPT and removed it. I think your comment here highlights a more glaring issue with the article: it does not mention RAG at all. I propose that some general information about RAG be added to the features section. Czarking0 (talk) 17:47, 15 January 2026 (UTC)[reply]
Google searched multiple things like "ChatGPT RAG" and I am not seeing good sources to add Czarking0 (talk) 17:53, 15 January 2026 (UTC)[reply]
The arxiv link is a primary source that I do not see the need to consider at this time. The AlJazeera article basically says PR firms hope that writing a bunch of promotional content can result in LLMs reading that content from web search results and repeating it. I don't really think that is something that needs to be added the article but I could be wrong. Czarking0 (talk) 18:00, 15 January 2026 (UTC)[reply]
As you mentioned above, there is nothing added from RAG, nor RAG poisoning. So I disagree with both statements here. I think the paragraph I propose is suitable to be appended under the FTC one. As we don't have consent, let's wait for other users to give their opinion. Josep a11 (talk) 23:03, 19 January 2026 (UTC)[reply]

Proposed merge of ChatGPT Search into ChatGPT

[edit]

Can and should be covered within the main article as this is just a feature of the software that the main article covers Czarking0 (talk) 17:54, 15 January 2026 (UTC)[reply]

On one side, the article ChatGPT Search remains a stub, and the media coverage of the feature has waned. On the other side, there are 14 other Wikipedia articles on ChatGPT Search in other languages, and the feature seems to me probably notable enough for a standalone article.
Some of the "History" section also seems too detailed to merge into the ChatGPT article, like the release history details and the sentence on the prototype.
So I lean toward keeping the separate article "ChatGPT Search", but potentially adding a sentence or two on that feature in the "ChatGPT" article if there is something important that is missing. Alenoach (talk) 00:07, 20 January 2026 (UTC)[reply]
Oppose - The ChatGPT page is already trying to cover a lot of headings and merging ChatGPT Search or anything else linked at Template:OpenAI series or Template:OpenAI (at time of writing, they list different products) wouldn't help matters. I have mixed feelings about the number of short ChatGPT-related articles there are, but I do not feel that this merge would be the solution. Here for the one billionth edit (talk) 15:05, 20 January 2026 (UTC)[reply]
  • Support merge: that's a feature, not a standalone product. There is no article about python sandbox used by ChatGPT, and this one is also completely unnecessary. Besides, it's a stub with nothing substantial, that can't be covered in one sentence in the main article. Artem.G (talk) 15:59, 20 January 2026 (UTC)[reply]
  • Support merge.This is a feature of Chat GPT and I do not see anythign to suggest that a whole article is warranted for this feature rather than being covered in the Chat GPT article (especially since as noted above other features do not have articles.) GothicGolem29 (Talk) 00:12, 21 January 2026 (UTC)[reply]

GPT Safety info

[edit]

There’s been a lot of talk on places like Reddit or Twitter about unfavorable answers given in response to harm and safety related topics, real or imagined. I believe this should be added to the article (assuming there are good sources for it; I can’t find any now) Miracusaurs (talk) 13:46, 16 January 2026 (UTC)[reply]

"Duplicated citations" template

[edit]

Hi GothicGolem29. I saw that you added back the "Duplicated citations" tag. I have explained here why these citations seem already ok in this case (one from an excerpt, the other one appearing in a grouped reference). I'm surprised that the tool that the DuplicateReferences script reports duplicated refs when one of the citation comes from an excerpt (and is thus not duplicated in the wikicode of the page itself). Having false positives due to excerpts isn't rare. Can this be improved in the script? I removed the other offending citation to avoid triggering the DuplicateReferences script again. Alenoach (talk) 02:18, 21 January 2026 (UTC)[reply]

Hello @Alenoach my bad for not responding to previous Comment i did not see an notification for that. The citation that you left is 121 https://www.cnbc.com/2024/05/13/openai-launches-new-ai-model-and-desktop-version-of-chatgpt.html and 141 https://www.cnbc.com/2024/05/13/openai-launches-new-ai-model-and-desktop-version-of-chatgpt.html and they are the same citation one listed in a table https://en.wikipedia.org/wiki/ChatGPT#Model_versions and one listed under 141 in the gpt 4 section. I am not sure why you think they are grouped together as they aren't and they arent really an excerpt they are the exact same article As I showed by copying thel ink from both citations and pasting them in this comment behind the number they are listed under in the citation.I do not see anything to suggest this sort of duplicated citation is allowed they should be listed as the same named citation as they are. GothicGolem29 (Talk) 11:11, 21 January 2026 (UTC)[reply]
Hi GothicGolem29, the second occurrence is within the "GPT-4o" subsection, which is an excerpt. You can see by searching the wikicode that the url is only visible for the first occurrence, the one in the table. The second occurrence is not in the wikicode, it's just transcluded from the article GPT-4o. It could be good for you to try to fix these duplications yourself, as it would remove the intermediate step of adding and removing the "Duplicated citations" template, save time for other editors, and you would get additional understanding of when and why reference deduplications can be impractical or infeasible.
When I mentioned grouped references, I was talking about the other one from MedPage Today, for which I removed the second occurrence to avoid triggering the DuplicateReferences script again. Alenoach (talk) 03:41, 22 January 2026 (UTC)[reply]
Hello. Thanks for clarifying though I am a little confused why you removed the second one to prevent the script triggering when you left the other one which would also trigger the script. Are you referring to the excerpt when you say infeasible? GothicGolem29 (Talk) 18:03, 22 January 2026 (UTC)[reply]
I guess you're right that even only one duplicated ref will trigger the script as well (I have never used that script so I trust you on that). And yes, when I say infeasible, I was referring to the CNBC source from the excerpt. As far as I know, it can't be deduplicated, but you can try if you want. Alenoach (talk) 19:16, 22 January 2026 (UTC)[reply]