Wikipedia:Articles for deletion/Alignment Research Center

The following discussion is an archived debate of the proposed deletion of the article below. Please do not modify it. Subsequent comments should be made on the appropriate discussion page (such as the article's talk page or in a deletion review). No further edits should be made to this page.

The result was no consensus‎. 78.26 _{(spin me / revolutions)} 23:35, 31 May 2023 (UTC)[reply]

Alignment Research Center[edit]

[Hide this box] New to Articles for deletion (AfD)? Read these primers!

(Find sources: Google (books · news · scholar · free images · WP refs) · FENS · JSTOR · TWL)

While it gets mentions, there is not enough in-depth coverage from independent, reliable, secondary sources to show that it passes WP:GNG. Onel5969 ^{TT me} 11:24, 7 May 2023 (UTC)[reply]

Note: This discussion has been included in the deletion sorting lists for the following topics: Organizations and California. Spiderone^{(Talk to Spider)} 11:30, 7 May 2023 (UTC)[reply]
Delete The notability case here is marginal at best, and if we were to have an article about the organization, we'd need to blow this up and start over. The line ARC tested GPT-4's capability to plan, replicate itself, acquire resources, remain hidden on a server, and conduct phishing attacks is a very close copy of the Ars Technica story. (Like the rest of the news I've looked at so far, that story doesn't cover the ARC as an organization in a way that the relevant notability guideline is met.) The sentence beginning ARC also found that GPT-4... is close enough to the Yahoo Finance story to count as copyvio. The line ARC has been expanding from theoretical work into empirical research, industry collaborations, and policy looks like it was written first and then had citations tacked on, one to the ARC's own website and the other to an online magazine of unknown reliability that mentions it in passing. Neither source supports the statement very well. In short: not covered as an organization, and the text would need a total rewrite to ensure it complies with policy, and indeed with basic academic ethics. XOR'easter (talk) 16:15, 7 May 2023 (UTC)[reply]
- Both of the close paraphrases have been corrected. If there are any others I am not aware of them. Sandizer (talk) 20:24, 7 May 2023 (UTC)[reply]
  Rewriting by making the passages less clear ("responded impermissibly"?) isn't really an improvement. XOR'easter (talk) 20:28, 7 May 2023 (UTC)[reply]
  I don't understand why that isn't more clear. Which is easier to understand, a response which isn't permitted, or a prompt which is forbidden? Clearly the latter (original) is misleading, because if a prompt requests forbidden information (the actual meaning here) then the correct response would be to decline to provide that information and explain why. Merely responding to such forbidden prompts isn't bad; failing to respond at all would be worse. Sandizer (talk) 20:47, 7 May 2023 (UTC)[reply]
Keep - as OpenAI's primary independent contractor for safety evaluations, this is a notably important organization. I have absolutely no connections to either. The argument here seems to be the depth of the coverage. The article currently has eight independent, secondary, reliable, mass media sources describing the organization as performing that functon for OpenAI, with most of them going into at least a paragraph of depth about the outcomes of that process. This is far more than than the single sentence mention shown as insufficient at WP:SIRS. Sandizer (talk) 20:05, 7 May 2023 (UTC)[reply]
A paragraph about a thing an organization did one time isn't reporting on it as an organization, which is what the relevant guideline requires. Nothing in the coverage provided so far suggests we need an article about the Alignment Research Center, as opposed to an article about GPT-4. XOR'easter (talk) 20:34, 7 May 2023 (UTC)[reply]
I don't understand. Why do you think the guideline specifies that reporting on the actions of an organization is not reporting on the organization? Perhaps you have heard the aphorisms, "actions speak louder than words," and, "it's not what you have, it's what you do with it"? And it's not just "a thing" we're talking about -- there are at least four distinct interesting results of the process summarized in the article so far. To be clear for those who might be reading this discussion without having looked at the article, we're talking about descriptions of those results in The Atlantic, The New York Times, Vox, Ars Technica, Vice News, and The Wall Street Journal, among other less prominent sources. Sandizer (talk) 20:47, 7 May 2023 (UTC)[reply]
Sources that describe only a specific topic related to an organization should not be regarded as providing significant coverage of that organization. Therefore, for example, an article on a product recall or a biography of a CEO is a significant coverage for the Wikipedia article on the product or the CEO, but not a significant coverage on the company. The situation here is analogous. XOR'easter (talk) 20:50, 7 May 2023 (UTC)[reply]
"Only a specific topic" is written in the singular. Currently the article summarizes how independent, reliable mainstream news sources have reported on:
ARC's mission;

the name of its primary founder;

the former company of its primary founder (qualifying him as "noted" and his position as "key", by the way);

its founding date;

its activities, and several distinct details thereof;

the GPT-4 assessment contract, including the general statements of work to be performed;

the result involving hiring a TaskRabbit worker for solving a CAPTCHA;

the quantitative result on forbidden responses;

the quantitative result on hallucinations; and

returning the FTX Foundation money.

There is no question that the sources cover several times more than "only a specific topic". As for the example given, a company doesn't perform a recall, a regulatory agency is the actor and the company merely complies to carry it out. In contrast, ARC performed the GPT-4 assessment work as a voluntary contract, the work and results of which represent multiple topics, each of which has been reported on by independent, reliable, mainstream news secondary sources. Sandizer (talk) 21:11, 7 May 2023 (UTC)[reply]
One assessment = one topic. XOR'easter (talk) 17:45, 20 May 2023 (UTC)[reply]

Relisted to generate a more thorough discussion and clearer consensus.
Please add new comments below this notice. Thanks, RL0919 (talk) 11:38, 14 May 2023 (UTC)[reply]

Keep: Agreed with Sandizer. CastJared (talk) 19:26, 14 May 2023 (UTC)[reply]

Relisted to generate a more thorough discussion and clearer consensus.
Please add new comments below this notice. Thanks, CycloneYoris ^talk! 09:25, 22 May 2023 (UTC)[reply]

The above discussion is preserved as an archive of the debate. Please do not modify it. Subsequent comments should be made on the appropriate discussion page (such as the article's talk page or in a deletion review). No further edits should be made to this page.