Talk:ChatGPT

Reads as a promotional article

Listing every new feature is not necessary for an encyclopedia. A lot of the article is also wholly uncritical and reads as a puff piece. TietoTeekkari (talk) 09:06, 18 July 2026 (UTC)

Thanks for the input Tieto, if you have specific changes in mind or sources you believe are due for included critical commentary then I'd love to include that. Czarking0 (talk) 14:56, 18 July 2026 (UTC)

@Czarking0, has this been resolved? Just going through quick fail criteria before I fully review the article. Wikiediter2029 (talk) Nosey? Stalk My edits 08:15, 27 July 2026 (UTC)

Negative phenomena widely linked to ChatGPT:

circular financing:

https://www.bloomberg.com/graphics/2026-ai-circular-deals/

suicide stories:

Deaths linked to chatbots

Chatbot psychosis:

Chatbot psychosis

The absurd energy cost:

https://spectrum.ieee.org/ai-energy-use

Extremely unpopular data centers:

https://news.gallup.com/poll/709772/americans-oppose-data-centers-area.aspx

And a host of other issues.

Most of all, the LLM technology is not accurately described. The text is written to suggest it has capabilities it does not have.

It should be made more clear that it is a next token predictor and not a reasoning engine. E.g. "Generates human-like text" is misleading and ambiguous.

And there are inserted marketing material like "sometimes at a level exceeding that of an average human test-taker" in the first paragraph of the Features section. TietoTeekkari (talk) 08:37, 27 July 2026 (UTC)

Hi Tieto, I don't think this article should be about everything "linked to ChatGPT". This article aims to be about ChatGPT itself. I reviewed the article after your first comment and was satisfied with the state of the article. Some of the information you mention either is or should be covered in OpenAI. The Chatbot psychosis issue has been extensively discussed in the talk pages please read the history before continuing that aspect of the discussion. Generally ChatGPT is one of the most viewed Wikipedia articles of all time, and a highly covered topic in the press therefore I do not expect to cover everything that ever mentioned ChatGPT. To determine what is due I believe we should look at what the best sources include.

I believe the description of the functioning of the LLM is verified by the sources and to a reasonable detail for a very broad audience. Describing KV-attention and multi-head systems would not be appropriate for this widely viewed article which focuses on a product rather than a more technical article like Large language model which focuses more on the technology. Czarking0 (talk) 14:58, 27 July 2026 (UTC)

Thanks for the input, you have helped me with reviewing. Wikiediter2029 (talk) Nosey? Stalk My edits 15:36, 28 July 2026 (UTC)

I will also place a tag where appropriate. Wikiediter2029 (talk) Nosey? Stalk My edits 15:39, 28 July 2026 (UTC)

If it's not supposed to include everything, that is at strong odds with the fact that features are described in extreme detail.

The fact that this is a well-read article means that the coverage should be neutral. At the moment it is written as a puff piece. That is not neutral. These topics I mentioned are _very_ strongly linked to ChatGPT, and should all be mentioned in the article.

The fact that the Reception section contains almost no negative reception is not good. Especially as most of the reception has been negative. Many, many polls have been made showing that general sentiment towards ChatGPT is negative. This is not reflected at all in the reception.

Re: description. The point is not to add detail, but to deacribe it less ambiguously. LLMs generate plausible text, not "human-like responses". That is not encyclopedic language, and it is misleading and something more at home in marketing material.

It makes me suspect that OpenAI is editing this article to make it seem more positive

P.S. my name is not Tieto, that is a simply one word in my nick. TietoTeekkari (talk) 21:44, 28 July 2026 (UTC)

The beginning of the "Reception" section is mostly positive (and representative the initial reaction to ChatGPT), but the rest is rather critical. For example, the subsections "In politics" and "In art" are very negative. More broadly, there is also a full section on "Limitations", and the third paragraph of the lead is fully critical. Not opposed to covering more criticism, but I think the current version isn't promotional, there is plenty of criticism throughout.

"human-like responses" is a pretty mainstream description that just describes the outputs without making assumptions on what happens inside the LLM, but if there are good alternatives I'm open to it. Note though that the objective of post-trained LLMs isn't just to generate text that is plausible. Next-token prediction is the pre-training objective, but post-trained LLMs are not just next-token predictors.

For the topics you mentioned, I could add a sentence or two about psychoses/suicide cases to ChatGPT#Sycophancy. The topic of circular financing, if notable, would be more relevant to mention in a company article like OpenAI. The topic of data centers opposition is also quite tangential, it's a general trend rather than something about ChatGPT, I don't know if it's relevant to mention.

The consumption of ChatGPT could be relevant to have a paragraph on. But it would require extra care: there are conflicting figures on the topic (some sources say 0.3 Wh per request, some others say 3 Wh or more), and it's difficult to be neutral because framing effects can completely shift the perception of how much is consumed. For example if you compare the consumption of typical ChatGPT requests to Google searches or LED bulb consumption, it may appear big; if you compare to household overall consumption, it may appear negligible. Also, the consumption depends a lot on the particular model, whether it's for generating text or images, whether you use something like voice mode or deep research, how much reasoning effort... Alenoach (talk) 02:31, 29 July 2026 (UTC)

Those seem to be relatively weak arguments not to improve the article by including common criticisms (and maybe removing irrelevant marketing-grade copy). Post-training is still aiming at plausibility. Even more so, as it's using human reviewers to filter out less plausible responses and boost more plausible responses.

ChatGPTs energy cost is well documented in peer-reviewed sources that have later been reported by notable news media: https://www.bbc.com/news/technology-67053139

There is still much, much more positive content in reception. And the criticism is largely superficial. It is not the criticism I hear every day from peers, nor is it the criticism I know a lot of people have. Here below are some big stories about how people receive AI in general. Why is this not reflected in the article about the product that started this whole thing?

In nearly every country polled, people are more concerned than excited by AI in general (the only exception is Israel):

https://www.pewresearch.org/global/2025/10/15/concern-and-excitement-about-ai/pg_2025-10-15_ai_2_01/

Americans oppose data centers for gen AI in general:

https://news.gallup.com/poll/709772/americans-oppose-data-centers-area.aspx

An OpenAI data center is strongly opposed by a large crowd:

https://thecurrentga.org/2026/07/23/openai-data-center-project-sparks-outrage-among-effingham-county-residents/ TietoTeekkari (talk) 08:41, 29 July 2026 (UTC)

Can you provide a quote for the BBC source on energy use? I read the source and I do not see a specific quantitative figure for the energy use that should be included in this article. Czarking0 (talk) 01:45, 30 July 2026 (UTC)

I don't think you need a specific value, but the energy cost should be there. I will take a look at the primary (peer reviewed) source for exact figures. TietoTeekkari (talk) 08:13, 30 July 2026 (UTC)

Fair enough, but please remember WP:SECONDARY. Czarking0 (talk) 14:31, 30 July 2026 (UTC)

Well I tried to make this but I am not sure how to put it in the article nicely.

10

20

30

40

50

60

70

80

90

100

U.S.

Italy

Australia

Brazil

Greece

Canada

UK

Argentina

Spain

Poland

Mexico

France

Netherlands

Hungary

Indonesia

Kenya

Sweden

S. Africa

Germany

Japan

Turkey

Nigeria

Israel

India

S. Korea

25-country median

•   More concerned than excited
•   Equally concerned and excited
•   More excited than concerned

Czarking0 (talk) 02:13, 30 July 2026 (UTC)

Ok well I converted to SVG and added to the article. I still hate this method. Czarking0 (talk) 14:55, 30 July 2026 (UTC)

IDK why it does not render here. Renders fine on my sandbox. Czarking0 (talk) 02:14, 30 July 2026 (UTC)

features are described in extreme detail I certainly aim to cover everything in wp:summary style is there something in particular you would suggest editing to improve this? Czarking0 (talk) 01:36, 30 July 2026 (UTC)

Mostly I just think we don't need to list as many features as have been listed here. E.g. plugins, Pulse, the phone feature, Deep Research all feel irrelevant. Some of the stuff under Language also reads as a listing. I also don't think prices need to be mentioned.

These are not at the level of detail I would expect from an encyclopedia. These details would be relevant in a sales listing, which is why I've talked about this resembling marketing material. TietoTeekkari (talk) 08:24, 30 July 2026 (UTC)

What you are describing though is not a level of detail but whether topics are covered at all. All these features are much more strongly related to ChatGPT than opposition to datacenters. ChatGPT is fundamentally a software product. The features are defining. Additionally, we do not determine what belongs in the article by what "feel[s] irrelevant". We apply WP:DUE. There is a plethora of sources connecting ChatGPT's features to ChatGPT itself. If you think we are covering those features in too much detail then I want to work towards wp:summary style but if you think they should not be included at all then I remain convinced they are due for inclusion. Czarking0 (talk) 14:25, 30 July 2026 (UTC)

I am using "feel" to not be nicer here. They are irrelevant. An encyclopedia is not the place to list minutiae about the service offering of ChatGPT.

Main features, of course. But I have mentioned the ones I believe are irrelevant. The main feature is the chat bot. Everyhing else is not so germane. The opposition is much more relevant than, say the "Pulse" feature. Or the other ones I listed above. TietoTeekkari (talk) 07:18, 1 August 2026 (UTC)

Deep research and plugins are worth mentioning to some extent. Pulse and the phone feature are more niche, and Pulse in particular is apparently getting replaced by "scheduled tasks" (source). I also find that some of the content of the "Languages" subsection seems a little too anecdotal.

On the broader point about the details of features, it's worth noting that the previous good article application was rejected after Uhoj argued that the article focused too much on controversies and not enough on explaining what ChatGPT is. So I would rather focus on incremental improvement in either direction by addressing specific comments as we do here than on an overhaul of the existing content. Alenoach (talk) 16:54, 2 August 2026 (UTC)

At the moment it is written as a puff piece. You keep saying this, but repeating it will not change the outcome. Please suggest specific changes and then we can collaborate. Czarking0 (talk) 01:37, 30 July 2026 (UTC)

I can write the specific edits myself, I am merely bringing it up in the talk page first so that you can understand where I'm coming from.

I would prefer not to split this discussion into several threads. TietoTeekkari (talk) 08:10, 30 July 2026 (UTC)

I recommend WP:BOLD and WP:BRD rather than the consensus building method you are using but thanks for your collaborative spirit Czarking0 (talk) 14:28, 30 July 2026 (UTC)

These topics I mentioned are _very_ strongly linked to ChatGPT, and should all be mentioned in the article. I disagree but if there is consensus then of course we will work together to make it happen Czarking0 (talk) 01:38, 30 July 2026 (UTC)

Many, many polls have been made showing that general sentiment towards ChatGPT is negative. This is not reflected at all in the reception. Ok add a citation to one? Czarking0 (talk) 01:39, 30 July 2026 (UTC)

I did cite you two polls and one news article about an OpenAI data center.

I would prefer if we don't split this into several threads. TietoTeekkari (talk) 08:08, 30 July 2026 (UTC)

Cite in the article I mean. I would prefer if we do split the conversation by topic. Czarking0 (talk) 14:25, 30 July 2026 (UTC)

Ok. Please be more clear when you discuss. Splitting it by topic makes this extremely hard to read during and after the discussion. Especially if this is the comment. "Ok, sure" does not need a separate thread. TietoTeekkari (talk) 07:21, 1 August 2026 (UTC)

What are you quoting? I crtl+f'd for your quote and I do not see it Czarking0 (talk) 07:23, 1 August 2026 (UTC)

It's a paraphrase of your comment, "Ok add a citation to one?". Or you can read it as an example of the type of comment I mean. I quoted it to make it clearly separate from the rest of the sentence.

This meta-discussion is only making this more difficult to read. Let's stay focused on the contents of the article. TietoTeekkari (talk) 07:30, 1 August 2026 (UTC)

It makes me suspect that OpenAI is editing this article to make it seem more positive That is entirely possible, but as you can see by this graph much of the page is written by Alenoach and I neither of whom work for OpenAI so I think we can jointly resolve these issues without casting aspersions. Czarking0 (talk) 01:42, 30 July 2026 (UTC)

Right. I am sorry for that. It simply reads a little one-sided to me. I should focus on that and speculate less. Did not mean to blame anyone. TietoTeekkari (talk) 08:12, 30 July 2026 (UTC)

GA review

The following discussion is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.

---

This review is transcluded from Talk:ChatGPT/GA3. The edit link for this section can be used to add comments to the review.

Nominator: Czarking0 (talk · contribs) 04:54, 27 May 2026 (UTC)

Reviewer: Wikiediter2029 (talk · contribs) 08:00, 27 July 2026 (UTC)

Comments

Hello, I will be reviewing this article today. Wikiediter2029 (talk) Nosey? Stalk My edits 08:00, 27 July 2026 (UTC)

After a look, there is issues with promotional issues as mentioned in the talk page, with a clean up tag added (due to discussion and not being fixed), hence a WP:GAFAIL, Quick Fail 3. Wikiediter2029 (talk) Nosey? Stalk My edits 15:58, 28 July 2026 (UTC)

Quick fail Check

No issues. Earwig says copyvio in lead but seems to be false after a check. Unsure if another issue is resolved. Wikiediter2029 (talk) Nosey? Stalk My edits 08:17, 27 July 2026 (UTC)

The discussion above is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.

Updates to model version status

Would one of the regulars here please review the edit request at Wikipedia:Requests for page protection/Edit#ChatGPT#Model versions? The request seems to be good faith but because this talk page is protected, the requester couldn't put it here. ~Anachronist (who / me) (talk) 15:00, 29 August 2026 (UTC)

I updated the table for o3, GPT-5.3 and GPT-5.4. I didn't make changes about GPT-5.6 because the secondary sources I checked didn't mention which version of GPT-5.6 is available with which subscription and because that kind of information has a tendency to get outdated. Also, the table intentionally doesn't have rows for sub-versions like Luna or Sol, to limit its size. Alenoach (talk) 22:20, 29 August 2026 (UTC)