Talk:Claude (language model)

? Mr Vili talk 01:24, 5 March 2024 (UTC)

I don't think Claude is sufficiently popular for now to justify separate articles. Unless there is a very significant media coverage in Claude 3 or future models, I would say no, keep it in one article. Alenoach (talk) 18:45, 5 March 2024 (UTC)

Sure. I think Claude 3 will have significant coverage compared to past models, but we'll give it some time to see Mr Vili talk 04:44, 6 March 2024 (UTC)

I suggest removing the template "Machine learning" ("Part of a series on Machine learning and data mining"). It looks too generic to me. But it's ok if you want to keep it. Alenoach (talk) 18:50, 5 March 2024 (UTC)

Actually, Claude does not appear in the series on machine learning, which is by the way primarily about machine learning techniques, not about products based on machine learning. So I removed it. Alenoach (talk) 00:35, 6 March 2024 (UTC)

claude is not available in all regions. 195.34.152.36 (talk) 13:15, 14 March 2024 (UTC)

There is a move discussion in progress on Talk:Claude-3.5 which affects this page. Please participate on that page and not in this talk page section. Thank you. —RMCD bot 01:51, 21 June 2024 (UTC)

Wondering if a layman would understand the opening section. Can we mention it’s an AI tool? I think people would grasp it easier if we made that more explicit in the opening. PacificPenman (talk) 15:06, 27 February 2025 (UTC)

I don't think this sentence saying that the Claude logo looks like an anus is warranted. The sourcing is not that strong. One source is from the New Scientist, but it just briefly mentions Claude among other chatbots. Another one is from a personal website, and the remaining one a YouTube video; all three are not specifically about Claude. I have not seen other significant sources about this on the internet. This is significantly less media coverage than when a new Claude feature comes out. Alenoach (talk) 21:19, 24 July 2025 (UTC)

I see where you're coming from, but I've seen a significant amount of references to it outside of the cited sources. It's interesting what you say about news coverage, because I had barely even heard of Claude before I came upon this story. –DMartin 05:23, 25 July 2025 (UTC)

The only article that I have seen covering the topic that isn't self-published is the one from New Scientist, which briefly mentions Claude alongside other chatbots. Do you have another source to propose? Alenoach (talk) 16:19, 25 July 2025 (UTC)

Looking around, I can see quite a few people discussing both in the blogosphere and on social media. I don't think it's necessary to cite every single one. In looking at WP:Humour it says " When a humorous work on a subject has attracted enough attention to be verifiable in reliable sources, it can be appropriate to reference that work on the page." I feel like this qualifies. –DMartin 18:36, 25 July 2025 (UTC)

I don't think this is important enough for an RfC, but I'd be interested to see what some other editors think, as we both seem like we won't be convinced. –DMartin 18:38, 25 July 2025 (UTC)

Yes, a RFC seems overkill. I propose requesting a third opinion on that page: WP:THIRD. Alenoach (talk) 21:56, 26 July 2025 (UTC)

Good idea(I didn't even know that was a thing). I've posted it there. `–DMartin 05:56, 27 July 2025 (UTC)

Response to third opinion request:

Most sources I can find attribute this as an overall trend in AI company logos, listing Claude as just an example. Likely undue for this article, however may be due in an article about AI as a whole. Coleisforeditor (talk) 14:20, 1 August 2025 (UTC)

The editnotice here is in the actual talk page, and nothing shows when I edit the page. Is this intentional, or do we need an admin to make it eally appear as an editnotice? Somepinkdude (talk) 00:50, 15 December 2025 (UTC)

If I understand correctly, for it to be displayed when trying to add a new comment to the talk page, a file Template:Editnotices/Page/Talk:Claude (language model) would need to be created, as explained in Wikipedia:Editnotice. For example, see Template:Editnotices/Page/Talk:ChatGPT, which contains the "Generative AI editnotice". The fact that it wasn't added isn't really intentional, it's just that there hasn't been many comments to remove in this talk page so it wasn't necessary. Let me know if I misunderstood your comment. Alenoach (talk) 01:18, 15 December 2025 (UTC)

I renamed ‘Claude Code’ to ‘Coding’ and included artifacts in that section but some of this categorization can be tricky and wanted to open it up for others to weigh-in and undo my edits if they think Claude Code is a better section title ~2026-59353-6 (talk) 01:42, 28 January 2026 (UTC)

Features like Claude Code are not strictly limited to coding, but I think that was a reasonable decision to rename the subsection as you did. Alenoach (talk) 05:55, 28 January 2026 (UTC)

Also, thanks for your work on this article! (debloating the "Models" section, improving the sourcing...) Alenoach (talk) 05:59, 28 January 2026 (UTC)

This is regarding this and this revert of @Alenoach:'s edits, but it applies in general.

These edits used vague and editorializing language. Articles should focus on specific, falsifiable claims without resorting to jargon or hype. We should not assume that readers know why Claude is any different from other models. Further, we shouldn't assume that it is any different from other models. That is what reliable sources are for. Our goal is to summarize those sources to provide context to readers. We cannot fill-in the gaps with vague filler or editorializing based on personal familiarity with the topic. This is a recurring issue with this article and this topic, likely due to WP:LLM misuse. Grayfell (talk) 23:28, 28 February 2026 (UTC)

The reason why many reliable sources mention Claude in the context of software development for each new release is because it's often considered marginally better than other models in this task, and so it is broadly used for that. Virtually any modern LLM has some degree of coding ability. So, saying it's "often used for software development" is marginally more informative and less vague than saying that it "can be used for software development". And it is no more jargony or complicated to understand either, if something is often used for software development, it also means that it can be used for software development.

"Particularly" is also not editorializing. You can use Claude without Claude Code for software development, so "particularly" is arguably a marginally more accurate word than "specifically".

For the other edit, where you removed "after the model requested it during a "retirement interview"", this is what both sources say. The term "request" isn't particularly anthropomorphism and is used in both sources, and I don't know how to phrase it in a way that satisfies you, so I didn't insist to avoid an editing conflict. I wasn't the person who added it, I just restored it because the paragraph is otherwise a little weird (why did they give Opus 3 a blog?) and because it was removed based on a flimsy rationale.

This also looks inaccurate or insufficiently sourced: "With the exception of models made for military use, Claude uses constitutional AI [...]". What the Time source says is "Models deployed to the U.S. military wouldn’t necessarily be trained on the same constitution, an Anthropic spokesperson said", so it says a potentially different constitution rather than no constitution (and it's unclear whether there will be any military model after the event with Pete Hegseth). Alenoach (talk) 00:43, 1 March 2026 (UTC)

Claude Code is trendy. If that's what "often" is supposed to convey, I don't think that's clear or neutral enough. "Often" is too broad in this context.

I'm not sure info about usage frquency could be included without a lot of context and attribution. I don't see that here, yet. The body of the article says it is "widely considered" the best, but it doesn't really say how often it's actually being used. (This is also a WP:WEASEL issue). The Wired source is based on a softball interview with the head of the company, so let's look at the Axios source. That source is intended to be a quick overview and contextualizes this as part of the vibe coding phenomenon. It also qualifies this by saying There's a big difference between being able to generate a working prototype via vibe coding and being able to create and iterate secure, enterprise-ready applications. So is that the source for it being used often? If not, which source is it, and why does that belong in the lead of the article?

Claude's "request" is a publicity stunt. Both sources describe this supposed request as coming from the company's exit interview (which is itself absurd). Engadget qualifies this supposed request with "apparently" a few times, while my reading of The Verge source is that Anthropic suggested this, and Claude "agreed". LLMs are designed to be agreeable, so this is too fishy to take at face value. This is just self-mythologizing from the company, and Wikipedia isn't a platform for PR.

I've removed the line about Military use, thanks for clarifying. Grayfell (talk) 01:34, 1 March 2026 (UTC)

Hello everyone,

I think we should remove a couple of sentences in Claude (language model). It states:

Context: "According to the Wall Street Journal, the U.S. military used Claude in its 2026 raid on Venezuela."

Unrelated sentence: "While it isn't known to what capacity Claude was used, the operation resulted in the deaths of 83 people, 2 of which were civilians, and the capture of President Nicolás Maduro."

I removed the unrelated sentence, but another editor undid it, and said that I should get consensus on the talk page before removing it. I don't think that it is necessary for every edit, especially when removing unrelated info. That said, I would appreciate your feedback and/or support.

Another unrelated sentence: "The action was described by John Colemean of the civil liberties group FIRE as a retaliatory First Amendment issue, specifically compelled speech."

I think these sentences would be better in 2026 United States Intervention in Venezuela. They don't seem to belong in an article about Anthropic's LLMs. These sentences are focusing attention on unrelated topics and should not be here. The Conservative Journalist (talk) 22:38, 24 March 2026 (UTC)

Discussing reverted edits is standard procedure, per Wikipedia:BOLD, revert, discuss cycle. I admit I'm not the best at following this myself, but it's the norm on Wikipedia. Not every edit needs to be discussed, but this one was reverted, so here we are. Since I'm not the one who added that wording in the first place, this talk page makes more sense than my own talk page. That's generally how WP:consensus works.

For convenience, here are the sources which are currently cited for the death toll:

• Christou, William (14 February 2026). "US military used Anthropic's AI model Claude in Venezuela raid, report says". The Guardian. Retrieved 14 February 2026.
• Amrith, Ramkumar; Hagey, Keach (13 February 2026). "Pentagon Used Anthropic's Claude in Maduro Venezuela Raid". The Wall Street Journal. Retrieved 14 February 2026.

Per The Guardian: The US raid on Venezuela involved bombing across the capital, Caracas, and the killing of 83 people, according to Venezuela’s defence ministry. Anthropic’s terms of use prohibit the use of Claude for violent ends, for the development of weapons or for conducting surveillance.

Per Al Jazeera: A total of 83 people, including 47 Venezuelan soldiers, were killed during the US special operation in Venezuela.

Per Axios: No Americans were killed in the raid. Cuba and Venezuela both said dozens of their soldiers and security personnel were killed.

Not all sources mention the deaths, but most sources explain that the the reason this matters at all is because Anthropic purports to be opposed to the use of AI for violence. Bombing and killing are forms of violence, so per the sources themselves, this is important context.

I have no objection to cutting the FIRE quote. I rephrased it to provide attribution and to avoid WP:WEASEL, but I don't think I did a great job and I don't think this is due weight regardless. Grayfell (talk) 03:28, 25 March 2026 (UTC)

Hello Grayfell, thank you for your reply. Good point on providing the casualty figures for why using Claude sticks out. I think the facts are not related to Claude (language model). We may place that it what used in this operation, but placing the specific details in this article just doesn't seem relevant. I'm not arguing with the well-sourced numbers or statistics, I'm just thinking we should not place the numbers in this article. A reader wanting to read about Claude will most likely feel that this is out of context; they are seeking to read about AI, not the details of the Venezuela raid. I think we should just say that it was used in the operation, linked to 2026 United States Intervention in Venezuela to read in more detail.

On the FIRE quote, I think we agree. That is definitely not needed in this article, because it is not relevant.

Thanks again for all the input. The Conservative Journalist (talk) 14:56, 25 March 2026 (UTC)

If this article were exclusively about the technical aspects of Claude as a language model, it would read more like a guidebook or fan-wiki than a general-audience encyclopedia. There are definitely readers looking for that, but that's not what Wikipedia is for. Instead, the article should also include information about how/where/when/why Claude is used in the real world, and that includes these kinds of consequences. While these deaths also contextualize Anthropic's stated goals, their public image, and their PR efforts as a company, it is also context about the model itself, per these cited sources. The US government used Claude, specifically. Therefor I disagree that this is out of context. Grayfell (talk) 20:54, 25 March 2026 (UTC)

Alright then, would you take out the FIRE quote and leave the rest? There would also be other editors who agree with me. And this statement is sort of negative on Claude and Anthropic, which might be classified as not neutral. I agree we should put that Claude was used in the operation; I only think that placing the very details and statistics is wholly unnecessary. This article is about artificial intelligience models, not about the raid. The Conservative Journalist 21:20, 25 March 2026 (UTC)

I've removed the FIRE line.

As for being negative, "neutral" in Wikipedia jargon doesn't mean it cannot be "negative". If reliable sources discuss facts which reflect negatively on Clause and Anthropic, a neutral summary of those sources will also reflect negatively. For us to intentionally avoid or downplay negative information for this reason would be a form of editorializing. Grayfell (talk) 21:44, 25 March 2026 (UTC)

OK, thanks. But placing details about another topic might be seen as not good editing. It's just not good editing. We know that the operation resulted in people's death, but it is unnecessary to place the minor details. Why would we have to put the aftermath of an event that Claude helped plan? The Conservative Journalist 22:32, 25 March 2026 (UTC)

We cannot assume that readers will know that the operation resulted in people's deaths. Is 83 deaths and the capture of a country's leader a 'minor detail'? That's subjective. Regardless, not every reader is American or Venezuelan, and even among those who are, not all are paying attention to US or Venezuelan politics. This context was provided by sources, so including a single sentence is appropriate. Grayfell (talk) 02:42, 26 March 2026 (UTC)

Wikipedia is an official encyclopedia which many people trust and use daily. This is not an encyclopedia where you mess around with sarcasm. The deaths could be irrelevant because we don't know how Claude was actually involved in the operation (e.g., shipping, planning, logistics, etc.). The statement is irrelevant in context because it does not have to do with Claude directly. We know it was involved in the operation, however, we don't need statistics in the Claude (language model) article. The Conservative Journalist 15:17, 26 March 2026 (UTC)

I don't know what an "official encyclopedia" means here, but I was not being sarcastic in this response. Per sources, these deaths were not minor details, and per sources, they directly explain why this incident was relevant to Claude as a model. Dismissing the death of 83 people as "statistics" is euphemistic, also. Grayfell (talk) 01:33, 27 March 2026 (UTC)

What is your point of stating the deaths and the capture of Venezuela's President? What are you trying to convey with this statement? I'm not trying to argue with the facts; they are quite accurate. I am trying to say, we may put that it was used in the event, but placing the number of deaths is not relevant. We have another article (2026 United States Intervention in Venezuela) for that. The Conservative Journalist 13:49, 27 March 2026 (UTC)

Since this was redundantly copied to multiple articles the first time around, I will copy/paste and edit my own talk page comments for why I removed it:

Regarding this revert from @Iljhgtn:, the sources for this being 'widely condemned' are two pages from Reason and one from Cato. Whether or not these are properly opinion articles is relevant, but they are definitely also not 'wide'. They are closely aligned and ideologically biased sources. Per WP:REASONMAG, that outlet's news coverage is reliable for statements of fact, but 'widely condemned' isn't supported by any one source as a simple fact. This is an opinionated source being used for a subjective claim without attribution in the lead. In order to say this was "widely condemned" we would need to cite and better summarize a reliable source saying that. To cite examples of condemnation and summarize them in this way is editorializing and WP:SYNTH. Likewise, "others have stated" is improper. Attributing direct quotes to organizations, instead of the named authors of those opinions, is also sloppy. Using a primary source for a quote to support a specific opinion is also undue, here.

Additionally, there were (at least) two notable amicus briefs, not just one, as the revert suggested. Both were from groups with specific, relevant ares of interest.

We also have an entire article on this at Anthropic–United States Department of Defense dispute. Reason, Cato, etc. are not mentioned there (yet). This article seems like a poor place to explain other party's opinions about this dispute, and this makes the editorializing and due weight issues stand out more here. Per above discussions, consensus seems to be against including this level of detail in this article.

An additional issue is WP:LEADFOLLOWS and cite-clutter. The cites in the lead are different from the cites in the body, and since the injunction isn't being contested, this isn't necessary in the lead at all. Grayfell (talk) 19:38, 7 April 2026 (UTC)

I also went over this one and removed "widely" from "widely condemned just now. Thanks @Grayfell. Iljhgtn (they/them · talk) 19:39, 7 April 2026 (UTC) WP:BE

Since Iljhgtn was blocked for sock puppetry, I've reverted. I welcome discussion from unblocked editors. Grayfell (talk) 01:51, 11 April 2026 (UTC)

Since Wikipedia isn't a catalog, is there a policy-based reason to include the 'knowledge cutoff' column in the version table? In other words, do reliable, independent sources regularly describe this detail as a defining trait in the same way as they describe the name, release date, and current status? Grayfell (talk) 22:20, 13 April 2026 (UTC)

The knowledge cutoff has become less important now that models typically have access to the internet, and it's rarely covered in independent sources, so removing it from the table is justifiable. Alenoach (talk) 09:20, 14 April 2026 (UTC)

Okay. I've removed it. Grayfell (talk) 19:00, 14 April 2026 (UTC)

The newest and pretty Popular gatekeeped Model isn’t mentioned in the Article yet. Claude Mythos is the biggest and smartest AI model we know of Time yet so it should be mentioned in the Article. ~2026-23022-02 (talk) 09:51, 14 April 2026 (UTC)

sorry i meant isnt explained enough yet ~2026-23022-02 (talk) 09:53, 14 April 2026 (UTC)

Wikipedia doesn't publish original research, and isn't a platform for promotion, so we need to summarize reliable sources. Anthropic isn't reliable for how smart a model is, so we would also need independent sources. Grayfell (talk) 18:56, 14 April 2026 (UTC)