Content deleted Content added
Assessment (Mid): banner shell, +Artificial Intelligence, +Computing (Rater) |
JoNeedsSleep (talk | contribs) |
||
| (12 intermediate revisions by 2 users not shown) | |||
Line 81:
<span class="wikied-assignment" style="font-size:85%;">— Assignment last updated by [[User:Yasmeenbg|Yasmeenbg]] ([[User talk:Yasmeenbg|talk]]) 21:35, 7 November 2024 (UTC)</span>
== Mechanistic Interpretability ==
I am planning to add a page about Mechanistic Interpretability since Mechanistic Interpretability is sufficiently well-defined as a field and wouldn't result in large overlap with the current article. Currently mech interp is a small subsection of this article - plan to expand it as a new article. Re the previous discussion about explainability vs. interp, I think the terms have changed enough to warrant clear distinction. [[User:JoNeedsSleep|JoNeedsSleep]] ([[User talk:JoNeedsSleep|talk]]) 18:45, 3 May 2025 (UTC)
:I would also say that the topic has become notable for a new article. Since you appear to be a beginner, I'll give you some advice: first, start by finding and reading [[Wikipedia:Reliable sources|reliable sources]] (preferably articles from mainstream media, for example ''[[Time (magazine)|Time]]'', although research articles published in peer-reviewed journals are also accepted; sources from blogs or YouTube may be rejected). Then synthesize what these sources say. Finally, write an introduction which acts as a summary of what the rest of the article says. [[User:Alenoach|Alenoach]] ([[User talk:Alenoach|talk]]) 19:18, 3 May 2025 (UTC)
::Thanks for the reply Alenoach! I'm planning to e.g. cite a blog post calling for more mech interp action written by Dario Amodei, CEO of Anthropic, as a primary source. Would that be considered a good source? [[User:JoNeedsSleep|JoNeedsSleep]] ([[User talk:JoNeedsSleep|talk]]) 19:29, 3 May 2025 (UTC)
:::When [[Wikipedia:Self-published sources|self-published sources]] are written by someone that is well-known rather than a random user, it gives them additional credibility, although it's still generally preferred to use sources from mainstream news websites. [[User:Alenoach|Alenoach]] ([[User talk:Alenoach|talk]]) 19:38, 3 May 2025 (UTC)
::::Got it, thanks. In terms of the logistics - do I add a redirect to the mech interp article in this xai article once I’m done? I don’t currently know what to do with the existing content about mech interp on this page. Appreciate your help. [[User:JoNeedsSleep|JoNeedsSleep]] ([[User talk:JoNeedsSleep|talk]]) 19:46, 3 May 2025 (UTC)
:::::Redirects, technically, are pages that only contain a link to another article. We could for example modify the redirect page "[https://en.wikipedia.org/w/index.php?title=Interpretability_(machine_learning)&redirect=no Interpretability (machine learning)]" to point to the new "Mechanistic interpretability" article instead of "Explainable artificial intelligence".
:::::But maybe you meant when you said redirect is "[[MOS:LINK|internal link]]". I guess the only thing that needs to be done in this article on xAI is to add either a [[Template:main article|main article template]] in the "[[Explainable artificial intelligence#Interpretability|Interpretability]]" subsection to the new article, or a simple internal link. [[User:Alenoach|Alenoach]] ([[User talk:Alenoach|talk]]) 19:58, 3 May 2025 (UTC)
::::::Hi Alenoach, thanks for the suggestion - that is my plan. I wrote the article here [[Talk:Mechanistic Interpretability]] but the url redirects to this page still. Do you have any advice for redirecting the mechanistic interpretability url to the right page? Thanks! [[User:JoNeedsSleep|JoNeedsSleep]] ([[User talk:JoNeedsSleep|talk]]) 03:42, 12 May 2025 (UTC)
:::::::Actually, I moved the redirect page. Would appreciate your feedback on [[Mechanistic Interpretability]]. On the point of redirecting Interpretability to Mech Interp, I think Mech Interp is historically a lot narrower than interpretability and would propose still redirecting it to xai, except with an expanded section. [[User:JoNeedsSleep|JoNeedsSleep]] ([[User talk:JoNeedsSleep|talk]]) 03:47, 12 May 2025 (UTC)
::::::::The article is well-written. I made [[Special:Diff/1290020387|some]] modifications. It would benefit though from having more sources from reliable news outlets or peer-reviewed journals, Wikipedia is pretty strict on which sources are [[Wikipedia:Reliable sources|considered reliable]].
::::::::For the redirect, are you sure you don't want the redirect to point to the article on mechanistic interpretability? I know mechanistic interpretability is more narrow than interpretability, but I suppose that's what people are interested in most of the time when they search for "Interpretability". [[User:Alenoach|Alenoach]] ([[User talk:Alenoach|talk]]) 09:23, 12 May 2025 (UTC)
:::::::::Thanks for the feedback, really appreciate it. I am not opposed to redirecting interp to mech interp. XAI is definitely not what people are looking for when they search interp and that interp section is way too brief, though my main concern is that I described mech interp more in its precise sense, and there’s non-“mechanistic” interpretability especially in various disconnected parts of academia. I do plan to incorporate more history of mech interp in academia in my next revision. Would you recommend I expand the scope of this article in my next revisiob? [[User:JoNeedsSleep|JoNeedsSleep]] ([[User talk:JoNeedsSleep|talk]]) 15:43, 12 May 2025 (UTC)
::::::::::Not necessary to expand the scope I would say, unless you think it would make the article better. Interpretability in general has a more fuzzy scope, and the overlap with explainability might confuse readers. I know redirecting from "Interpretability (machine learning)" to "Mechanistic interpretability" is not an exact match, but I suppose the link to "Mechanistic interpretability" would still be more useful, so I have a slight preference for redirecting to that. [[User:Alenoach|Alenoach]] ([[User talk:Alenoach|talk]]) 15:56, 12 May 2025 (UTC)
:::::::::::That makes sense, I'm with you on that judgment. [[User:JoNeedsSleep|JoNeedsSleep]] ([[User talk:JoNeedsSleep|talk]]) 16:32, 12 May 2025 (UTC)
| |||