Wikipedia typically summarizes the cited source. LLMs don't use the language in the source to write a summary. It comes up with its own language. LLM cited text usually gets removed for uncited claims because the claims just aren't in the reliable source.
The best use of AI is when it can look up existing text that you know already exists and apply it to your circumstances.
Wikipedia is the opposite of that. It is the existing text. AI Wikipedia would be the exact ouroburos of bullshitifying the Internet that experts have been warning about.
It could have some tangential uses, done responsibly. "Hey Claude, check the source material for all of these citations and find any that disagree with the way they're being used in the Wikipedia article."
And then, critically, you have a human review that output.
The temptation to skim and feel like you've done all the hard work in looking it up and synthesizing after only just hitting the lazy button is wild. I use deep research functions a lot for work and I was super naive in my ability to grasp the underlying knowledge off reading that content let alone trusting it. Found my knowledge was super squishy or lacking depth to answer any questions or leverage it with any meaningful degree that i normally would have prior to this tech.
Its been embarrassingly harder than I would like to admit in trying prevent myself from using it in that way other than just a fancy google search. The temptation is there like fast food ready to hit my veins - but I am getting better
Yep. I’ve been using AI more for work and if I’m not vigilant it will fuck up. It’s like having a very fast intern. You still have to check their work carefully.
I’m the same. AI search is usually my first port of call these days because traditional search engines are so shit now. I’d estimate that its summaries are around 80% accurate. Yet, even knowing that, i still have to fight the temptation to just accept what it says and instead check the sources
I got certified for a software suite and at the end during Q & A the instructor just pulled up chat gpt with all the install manuals preloaded to answer people's hardware/version specific questions.
Everyone acting like AI being useful is an outlier is coping. Rightfully so given the state of the world but still coping.
The hype among corpos that AI is going to replace all labor is stupid. But when the bust happens and the hype dies down, we're still gonna be ordering pizzas online, Amazon will exist, and it'll be a thing we work with forever.
What's different about this boom is that it's being used to fund purchases of land, computer hardware, and data centers. Unlike the dot com bust, where all the devalued companies left behind was mostly useless websites.
So they'll be able to maintain their market dominance of AI, just under private ownership instead of the defunct public companies that went bust.
This is a ball and cup game, more so a Ponzi scheme, than the fomo pump and dump that was dot com.
Uhhh, ask me anything, I guess? This is the first I'm hearing of this guideline, as I missed the RFC, but it was published February 10, so I'm confused why 404 is suddenly reporting it now (edit: I was confusing it with the LLM translation guideline which they also link). 404 gets a subtle nuance wrong by calling it a "policy" rather than a "guideline", which are technically different on Wikipedia, but it's not worth splitting hairs for the general public.
Wikipedia editor, Ilyas Lebleu, who goes by Chaotic Enby on Wikipedia and who proposed the guideline
I've actually talked to them a few times before; they're really cool.
Edit: Oh, okay, the article is actually discussing two guidelines. This one is what they're mainly referring to. Speedy deletion criterion G15, a precursor to this, is related and says that articles predominantly created by an LLM without human oversight can be speedily deleted.
all 23 comments