Talk:ChatGPT

Currently the ChatGPT Apps section relies only on primary sources. The original edit that introduced it is tagged as possibly AI generated. I think overall this section is promotional and undue weight given a lack of secondary sources covering it with the same caliber that much of the rest of the article has. I plan to remove it entirely unless good secondary sources show due weight. Czarking0 (talk) 15:58, 15 April 2026 (UTC)

There could be a sentence or two on ChatGPT Apps (whether or not in a dedicated subsection), but the current version seems too detailed and a little promotional.

By the way, I think the section "Lawsuits" should be moved to the article on OpenAI. Alenoach (talk) 08:10, 16 April 2026 (UTC)

An article is meant to explain what its subject is. This one falls flat. Rather, much weight is given to criticism of chatGPT, of OpenAI and of AI writ large. Compare with the content of Gemini, Claude, Llama, etc.

Instead of endorsing the many criticisms, we could summarize them in a Criticism section that explains that this product / company / CEO is the subject of withering rebukes and describes the critical views. Alternatively, we could split this stuff off to Criticism of chatGPT and Ethics of artificial intelligence.

Specific problems include:

• "The laborers were exposed to toxic and traumatic content; one worker described the assignment as "torture"."
  • This is a criticism of content moderation in general. To imply that it's unique to chatGPT is misleading.
• "ChatGPT's training data only covers a period up to the cut-off date, so it lacks knowledge of recent events."
  • This is a problem for LLM technology in general; not chatGPT in particular.
• "Training data also suffers from algorithmic bias."
  • Again, a general problem.
• ChatGPT#Hallucination insinuates that chatGPT is somehow worse than other other implementations in this regard. If that's verifiable we should say it directly.
  • Ditto for ChatGPT#Jailbreaking.
• ChatGPT#In_art has a deceptive heading given that the content is criticism and stuff that's not about art.
• ChatGPT#Lawsuits needs better sourcing to show that this merits weight.

Uhoj (talk) 17:28, 16 April 2026 (UTC)

In general I think you provide good bad criticism and I want to take some of this into account. However, in regards to a criticism section I agree with WP:CRITS. Criticisms being true of other systems is not a policy based reason to not include them here. We should follow the sources and include general criticisms when sources make them about ChatGPT in particular. Czarking0 (talk) 00:58, 17 April 2026 (UTC)

Having a criticism section or article is often better to avoid, as explained in WP:CRITS.

Things like hallucination, jailbreaking and bias are indeed not specific to ChatGPT, though media coverage attributed it in particular to ChatGPT at the time, because it was the first widely-used chatbot, so it got a lot of praise and criticism. I'm not sure if it's acceptable to keep it or not. For jailbreaks in particular, things like DAN seem very unlikely to still work; since the section title suggests that these are current limitations, I would lean toward removing it. Nowadays, sycophancy is more of an issue than jailbreaks.

For ChatGPT#In_art, you're right, I will check if there is a good way to reroganize it to address that problem.

For ChatGPT#Lawsuits, the current content about the stalking lawsuit is relatively niche, but we could add content about more high-profile lawsuits if we choose to expand it. It could be a little similar to OpenAI#Governance_and_legal_issues but with more of a focus on ChatGPT's behavior. Or we could perhaps just move everything to the article on OpenAI, or delete it if this particular lawsuit is not sufficiently notable. Alenoach (talk) 05:08, 17 April 2026 (UTC)

I agree with the others that WP:CSECTION applies. Sometimes the raw quantity of critical content would overwhelm an article, which does justify a split (examples include Criticism of Google, Criticism of Facebook, etc.) but there are two issues with that. The first is that it would be premature here. The second is that such an article is not an excuse to sanitize the main article. We have to follow sources, and any source which intentionally avoids valid criticism is inherently less reliable.

Incidentally, if you know of reliable sources which are critical of those other models and which aren't already cited, please post them to Talk:Gemini (language model), Talk:Claude (language model), Talk:Llama (language model) etc. Grayfell (talk) 05:30, 17 April 2026 (UTC)

I've trimmed the jailbreak section, I think that this has appropriate weight, and the links are LLM general.

See what you think. I'll see if I can improve some of your other concern areas. WhaleFarm (talk) 01:07, 18 April 2026 (UTC)

I made some adjustment to the jailbreak section, I think DAN is more notable to cover. Let me know if it's not ok. I also modified the content on knowledge cutoff dates to clarify that it's a problem with all LLMs and to condense it. Otherwise, I think the consensus is that there isn't really a POV issue, so I suggest removing the POV template (though we can continue to improve it if we have ideas). I believe the remaining issue is what to do with the "Lawsuits" section. Alenoach (talk) 10:28, 18 April 2026 (UTC)

I read the citation (msn) cited in the ChatGPT#Lawsuits section, and I would agree, using only that, that the article gives it too much weight. However, the msn article is really only a pointer into the techcrunch article, which is very well written, and would support at least this much weight. I say this as no fan of techcrunch, which rarely has much to add beyond simple copying.@Uhoj, would you look at and opine? WhaleFarm (talk) 02:31, 18 April 2026 (UTC)

I moved the content of the Lawsuits section to the article on OpenAI, because it was too focused on one particular lawsuit, not mentioning other notable ones (such as Raine v. OpenAI or, the NYT lawsuit or the one about the murder of Suzanne Adams). We could potentially have a "Lawsuits" section in the article on ChatGPT that is a summary of those in OpenAI#Governance_and_legal_issues with a focus on ChatGPT and its behavior; I let others decide whether we should do it or not. Alenoach (talk) 14:42, 18 April 2026 (UTC)

I think that's the right call, technically, no one is suing chatgpt. But maybe there should be a chapgpt/llm heading for "behavior attracting lawsuits." I'm sure somebody can come up with a better title, but that's the idea. WhaleFarm (talk) 15:41, 18 April 2026 (UTC)

Perhaps a "Sycophancy" subsection in the "Limitations" section. GPT-4o's sycophancy (telling people what they want to hear even when it departs from reality) is a major culprit behind both the murder of Suzanne Adams lawsuit and the stalking lawsuit. Alenoach (talk) 15:50, 18 April 2026 (UTC)

I think that's appropriate, do you want to do that now, or wait until after it's been stable for a while? What does @Uoj think? If it stays stable a while, I'll proceed with the GA review. WhaleFarm (talk) 15:55, 18 April 2026 (UTC)

I will start drafting something and will publish it if the quality looks good enough. Alenoach (talk) 16:09, 18 April 2026 (UTC)

Point two of the criticism essay says:

Do not present the material in a way that emphasizes beyond the emphasis given in reliable sources.

That's essentially a restatement of our policy on balancing aspects, so it's worth looking at what's emphasized in other overviews of this topic.

As a first step I propose merging Applications into Reception.

Britannica's take

• Software that responds to questions using natural language
• Disturbing to academics and journalists because it generates text that passes as human
• Based on a probability model constructed by processing large quantities of existing text
• Underlying technology is neural networks
• Was interpreted as a dramatic advance in AI when introduced
• User base grew rapidly
• Concern that students could use it to cheat
• Good at written composition; bad at math
• Hallucinates like other LLMs
• Fails to admit when it doesn't know an answer

IBM's take

• Was met with enthusiasm
• The first positive experience that most laypersons had with AI
• Democratized access to AI and the ability to create computer programs
• Provides a way for humans and machines to collaborate
• Started as an LLM, but has become multi-modal
• Chain-of-thought was an important advance
• Small models have some advantages
• AI and human thought are complementary
• There's a commercial race afoot
  --Uhoj (talk) 20:07, 17 April 2026 (UTC)

It sounds like the deeper issue is that a lot of this belongs at chatbot instead. But... if sources about ChatGPT specifically are critical, the article shouldn't intentionally downplay those sources by tucking it away in some other article. That said, 'applications' and 'reception' are two different things, so I'm confused about this proposal.

Oh, that raises another issue: 'applications' is likely a confusing section title if ChatGPT has its own app store, so maybe that should be changed either way.

As for the sources, Britannica makes sense for comparison, but the IBM one doesn't seem useful. That article is mostly just the personal opinions of Shobhit Varshney and Chris Hay, and it's not an impartial source here.

Grayfell (talk) 03:20, 18 April 2026 (UTC)

Do not merge two top level sections without consensus. I do not support combining the Applications and Reception section and I largely disagree with your conclusions about how to improve this article. I think this discussion should return more to the realm of citing guidelines that support our rational.

Content belonging on a different page does not indicate that it does not belong on this one. Wikipedia supports duplicate content directly with tools like Template:Excerpt which has been used on this page multiple times. Near copies without excerpt are also common. I use both extensively in WP:WikiProject Colorado.

While I find metaencyclopedic comparison interesting if an idea for this that is good will be good regardless of if Britannica agrees. Czarking0 (talk) 01:00, 19 April 2026 (UTC)

Why did you remove the information that I added about what people use ChatGPT for? The edit summary says sourcing does not show due weight for a top level section. This gives a reason for why the heading was removed. It provides no reason for removal of the content.

I do not understand what you mean by discussion should return more to the realm of citing guidelines that support our rational. I argued that WP:CRIT and WP:NPOVD support the notion that changes are needed. I don't think those arguments have been rebutted.

Regarding Content belonging on a different page does not indicate that it does not belong on this one this is not an argument that I've made. Uhoj (talk) 01:27, 19 April 2026 (UTC)

The content of Applications is not actually about applications. Instead, it covers the same sort of ground as Reception. Each subheading of Applications covers reception in a particular area. A section that explains what people use this thing for is a good idea in the abstract, but we don't presently have content to fill such a section.

While each fact mentioned in the article might be presented fairly, the very selection (and omission) of facts can make an article biased.

Compare our take with that of Britannica or IBM.

Our take

• Training
  • RLHF
  • Underpaid Kenyans being tortured
• Features
  • List of applications
  • List of release dates
  • Versions
  • Cost
  • Contested statement about water usage presented as fact
  • List of Features
  • Advertising
• Limitations
  • Limitations known to be generic to AI technology
  • A security breach
  • One guy losing his data
  • A watermarking feature that was never implemented
  • Content moderation
• List of versions
• Reception
  • An intro that covers reception
  • A guy saying he doesn't like it
  • A politician saying the end is nigh
  • Accusations of bias
  • A boycott movement
  • China, Italy and Russia
  • OpenAI restricting access and pushing back against sovereigns
  • A short-lived ban by Italy
  • A suit against openAI
  • Opinions of tech titans
  • Copyright concerns that are generic to AI. Fails to mention context that Anthropic got in hotter water.
• Applications
  • A laundry list of stuff that's hard to distinguish from the Reception section.

-- Uhoj (talk) 19:09, 18 April 2026 (UTC)

To me, this list summary is too opinionated to read like a neutral summary of the actual article, so it undermines your argument. Also, as I said above, the IBM source isn't useful for comparison.

The first sources you've cited here is for a 'contested' statement, but that source doesn't contest the single sentence paragraph in the article, it just says "it's more complicated than that" about a different source which isn't cited in this article. The second is specifically about a completely different company and doesn't directly mention 'ChatGPT' at all.

Since you don't have consensus, and these multiple tall lists of bullet points don't seem to be persuading anyone, I think you should find a different approach. The boilerplate advice in situations like this is to focus on specific, actionable 'change X to Y' type proposals to demonstrate the issue. Otherwise, Wikipedia isn't the place to promote a company's product line. Grayfell (talk) 02:22, 19 April 2026 (UTC)

Oh, also, there are multiple problems with this edit. In addition to being redundant with the 'applications' section, per NBER.org, this is a working paper and falls under WP:PREPRINT. Citing it for its own subsection with three paragraphs is not going to work. It's also noteworthy that six of the working paper's seven authors are directly affiliated with OpenAI, which is another point against its reliability here. If this has been published somewhere, it might be usable with context/attribution, but there's no obligation to cite something like that. Grayfell (talk) 02:35, 19 April 2026 (UTC)