Talk:Claude (AI)

Splitting proposal

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.

---

We should discuss whether Claude is primarily an LLM or a chatbot.

Sources describe it as both a chatbot and an LLM. Confusion arose because Claude launched its chatbot and first LLM under the same name, i.e., Claude (unlike ChatGPT). Recently, models have had their own names, like Mythos, Sonnet, Opus, etc .

Based on the consensus, the article should be renamed to either:

Claude (Chatbot)

Claude (AI), based on https://claude.ai

Claude is a consumer-facing app used by millions (available on desktop and phones), and I think that is what most people would think of when they hear "Claude." Claude models are now referred to by their specific model names rather than simply "Claude."

Since this article talks about models and their products, I support renaming it to Claude (AI). 🐈Cinaroot   08:24, 14 May 2026 (UTC)

It's two separate things with the same name, like Google Gemini and Gemini (language model). Claude the language model is to Claude the chatbot what iOS is to an iPhone.

We could probably split this article into two like the articles for Gemini. One article (Claude (chatbot)) would cover the chatbot itself, while the other (Claude (language model)) would cover the individual versions of the LLM (Opus, Sonnet, Haiku, Mythos, etc.) SuperPianoMan9167 (talk) 19:38, 14 May 2026 (UTC)

Yeah - much better idea. 🐈Cinaroot   20:37, 14 May 2026 (UTC)

Oppose the two already work fine together in the same article. Splitting may cause some confusion. Rager7 (talk) 04:37, 11 June 2026 (UTC)

Oppose - splitting would significantly increase the maintenance work, and there would likely be a lot of overlap between the two articles. Moreover, while the distinction between LLM and chatbot may be clear to us, it is not for most of the readers who land on this page trying to learn what Claude is. For them, having one comprehensive, better-maintained article is probably better. Alenoach (talk) 04:19, 15 May 2026 (UTC)

I think you have a point. Would you support a rename to Claude AI? 🐈Cinaroot   04:57, 15 May 2026 (UTC)

I think the fact that it's a vague name may actually be a good thing here, because it's sufficiently broad to cover the LLMs, the chatbot and other user interfaces or agents. And it's very concise. "Claude (AI)" is probably a little better since "Claude AI" isn't an official name. I would support a renaming to "Claude (AI)". "Claude (Anthropic)" could also be an option. Alenoach (talk) 05:36, 15 May 2026 (UTC)

Yes, Claude (AI) is my original proposal 🐈Cinaroot   06:36, 15 May 2026 (UTC)

• Oppose split, per Alenoach. The article is already a pain to maintain, if we're going to create more work, it would have to be for a much better reason. As for a rename, I oppose "Claude AI" as that isn't the WP:COMMONNAME, but "Claude (AI)" has potential. Grayfell (talk) 06:14, 15 May 2026 (UTC)

  I'm not entirely sure if it's a good idea to use "AI" with no qualifiers because that's a very common marketing buzzspeak practice that conflates "LLM" with "AI". See User:Altoids0/Call them LLMs. SuperPianoMan9167 (talk) 12:40, 15 May 2026 (UTC)

Good point. Grayfell (talk) 18:37, 15 May 2026 (UTC)

https://www.merriam-webster.com/dictionary/ai

https://www.dictionary.com/browse/ai?q=AI

https://www.oed.com/search/dictionary/?scope=Entries&q=ai

AI is Artificial intelligence. According to oxford - The earliest known use of the noun AI is in the 1960s.

It’s not a marketing term. Even if it is - nothing wrong with that. It’s in the English dictionary. Now we have a new word. People create new words. Words evolve. Now it’s part of people's vocabulary. Its not collusion like Altoids0 is claiming 🐈Cinaroot   19:49, 15 May 2026 (UTC)

The point that Altoids0 is making is that "AI" refers to a broad field of science and engineering characterized by the demonstration of intelligent behavior by machines, not "a program that uses neural networks to generate stuff". Not all AI is generative AI. Generative AI is just the most popular right now by far. The reason that Merriam-Webster even includes the definition a program or set of programs developed using tools (such as machine learning and neural networks) and used to generate content, analyze complex patterns (as in speech or digital images), or automate complex tasks is because ChatGPT and LLMs in general have "sucked all the air out of the room". SuperPianoMan9167 (talk) 19:56, 15 May 2026 (UTC)

Using "AI" as a synonym for "ChatGPT" or "LLM-powered chatbots" is marketing. SuperPianoMan9167 (talk) 19:57, 15 May 2026 (UTC)

Understood. Claude (GenAI) maybe a better option then. But AI is just a common name for GenAI - I don’t mind using it. 🐈Cinaroot   20:45, 15 May 2026 (UTC)

Yeah, Claude (AI) is better because of WP:NATURAL and WP:CONCISE. SuperPianoMan9167 (talk) 21:04, 15 May 2026 (UTC)

I also oppose the split, and furthermore think that Claude (AI) makes more sense if the split is not to occur, since now we are combining both the chatbot and the model information, and AI is a good enough umbrella term for both. Drivingdownthe401 (talk) 20:19, 18 May 2026 (UTC)

Agreed, using the term "AI" is fine. Arguably, other terms like "GenAI" or "LLM" sound more industry-friendly, but avoiding any such connotation should not be the focus of the discussion, let's refocus on finding a clear name that describes well the content of the article. Alenoach (talk) 21:15, 15 May 2026 (UTC)

Hi Cinaroot. The discussion Talk:Claude_(language_model)#Splitting_proposal is getting stale, and people are mostly opposed to splitting the article. But there is probably enough support for a renaming to "Claude (AI)". Maybe we should open a formal move request (with the {{subst:Requested move}})? Alenoach (talk) 19:59, 31 May 2026 (UTC)

I requested to close the proposal. I think the closer can move the page from this split proposal. I don’t want to waste people’s time with a new requested move. We have consensus to move the page 🐈Cinaroot   21:09, 31 May 2026 (UTC)

Seems reasonable. Is the request still pending? I don't see it in Wikipedia:Requested moves/Technical requests, so I don't know where it was done. Alenoach (talk) 22:09, 11 June 2026 (UTC)

In the meantime, I propose to remove the "Split" template: there is a clear consensus against a split, and the debate isn't about splitting any more. And if we are making a technical request, we don't need a "Requested move notice" template. Alenoach (talk) 22:19, 11 June 2026 (UTC)

Alenoach Closer comment: it's pending at WP:CR. I'm waiting for the discussion to stabilize before taking action. In solidarity, Iseult ^Δx talk to me 06:41, 12 June 2026 (UTC)

Support move to Claude (AI), oppose split. Wasting time is still my passion (talk) 05:32, 15 June 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.

Proposal: Add harness engineering and loop engineering to Claude Code section

I'd like to add a paragraph to the Claude Code section describing the architectural concepts revealed by the March 2026 source code leak. The article currently mentions the leak (refs 28-29) but doesn't describe what it revealed about Claude Code's architecture.

Proposed addition after the leak mention:

"Analysis of the leaked source code revealed that approximately 98.4% of the codebase consists of operational infrastructure—the 'harness' that manages tool execution, permissions, context, and memory—while only 1.6% constitutes AI decision logic. This architecture separates the model from its execution environment, with the harness enforcing safety boundaries and enabling autonomous operation. The leak also exposed primitives for autonomous execution loops, reflecting a shift toward what practitioners have termed 'loop engineering'—designing systems that schedule, verify, and persist agent operations without continuous human prompting."

Sources:

• arXiv 2604.14228v1 — "The Design Space of Today's and Future AI Agent Systems" (VILA Lab, MBZUAI and UCL, April 2026). Analyzes the leaked architecture, provides the 98.4% harness / 1.6% model statistic, and formally defines the harness concept.
• The New Stack (June 12, 2026) — "The Anthropic leader who built Claude Code says he ditched prompting — now he just writes loops." Covers Boris Cherny's statement and the "loop engineering" term.
• Existing refs 28-29 (Ars Technica, CNBC) provide the leak context.

This addition connects the existing leak coverage to the architectural insights it revealed, using peer-reviewed and established tech publication sources. CharlesLiuson (talk) 01:52, 14 June 2026 (UTC)

Hi, CharlesLiuson. The sourcing for the technical details looks quite thin (the arXiv paper is a preprint, so it shouldn't be used here), and I personally find it interesting, but I'm not sure most of the audience would understand what's the difference between operational infrastructure code and AI decision logic code.

The source from The New Stack looks ok to use. It doesn't mention the leak, but there could be a sentence or two on loop engineering, here or in another article like Prompt engineering. Also, just in case, note that if you add content, it should be written in your own words, not AI-generated (see WP:LLM). Alenoach (talk) 04:44, 14 June 2026 (UTC)

While I agree with the gist of what Alenoach is saying, I don't accept that the New Stack article qualifies as a reliable source. I don't see sufficient indication of editorial oversight, fact-checking, corrections, etc. which are required for a source to be reliable. It appears to be a platform which accepts contributor content, and is therefor WP:UGC. Further, that platform is a subsidiary of Insight Partners, which has a conflict of interest with Anthropic. Grayfell (talk) 05:16, 14 June 2026 (UTC)

Claude code is a subscription service

Is this disputed as one of the defining characteristics, why remove it from the article? Czarking0 (talk) 14:17, 1 July 2026 (UTC)

But why is it a defining characteristic? There is no separate Claude Code subscription, it's included in Pro and Max, same as Claude Cowork, Claude Design, etc. I don't think adding "a subscription service" to all three is necessary, but maybe a sentence or two about free vs paid versions are needed. Artem.G (talk) 15:01, 1 July 2026 (UTC)

I agree no need to constant rehash how to pay this is not a catalog. I plan to treat payment on this article similar to how it is treated on ChatGPT. I'll drop the point about defining characteristic if you disagree which I think is ridiculous but I do not need to litigate it. It is however due for inclusion based on the sources. Czarking0 (talk) 03:50, 2 July 2026 (UTC)

Necessity of the “user base” section

ChatGPT has cultivated similar sects of users with abnormal attachments to its software, such as the audience of users who began dating their versions of the GPT-4o model. Is it appropriate to imply Claude users to have an affinity for their software that is distinct from others? AstronautPreference (talk) 09:53, 10 August 2026 (UTC)