Post
![](data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="128" height="128"><rect fill-opacity="0"/></svg>)
Eugene Vinitsky 🍒
‪@eugenevinitsky.bsky.social‬
Some of the anti-AI stuff feels a bit like when people would say "don't use Wikipedia as a source." It's just like anything else, a piece of information that you weigh against multiple sources and your own understanding of its likely failure modes
April 26, 2025 at 6:23 AM
37 reposts
438 quotes
436 likes
Will you go awry if you use it as ground truth? Yeah! But that's a horrible idea for any information source, not just AI
The point is that students don't know that as their education advances they need to stop looking to elementary sources like encyclopedias & begin using primary & more in depth secondary sources unless they are told. Also encyclopedias cite to sources, LLMs don't.
this is the most wrong anyone has ever been. shut the fuck up
Unlike LLM outputs, articles are vetted and edited. While there is bias in what topics are covered, and the depth, for scientific topics, Wikipedia is a reliable place to start. www.nature.com/articles/438...
That's... true, but not meaningful.
Sources can be extremely trustworthy, batshit crazy, or anywhere in between. Saying that you should verify your sources does nothing to place an individual source on that spectrum.
Is AI as accurate as Wikipedia? No.
1/2
And that's before we weigh environmental impact, intellectual property theft, etc.
So the part I agree with you on is "If you use AI, you'll need to fact check everything."
But also, you could not use AI, and it would save you time.
2/2
Right but data retrieval isn't the primary use of this form of AI. You'd use it more for things like summary, synthesis, styling or translation
The research models will find and pull in web links and also do that. So OP's point is "understand the limitations of the tool" 🤷‍♂️
More
Wikipedia has citations that can be reviewed. The output of LLMs is a slurry from numerous sources.
To the extent LLMs cite "sources," they're often made-up.
You can follow wikipedia sources and find out for yourself.
Wikipedia contains actual information, put there by actual humans, who spend a lot of effort on verifying its accuracy.
An LLM generates sentences based on a statistical model of language, and has no concept of "information", let alone "accuracy".
Hope that helps!
Wikipedia is edited and maintained by real people with real sources and real facts.
I asked ChatGPT to cite 3 sources for a paper about bluebirds ... & every single one was imaginary. So no, I'm not going to trust its "facts".
That is correct and proper use of the tool
LLMs are better for dynamic functions like summary, synthesis, styling or translation
what a weird comparison on two entirely distinct and unrelated topics.
unrelated? didn't he explain pretty clearly how they're related?
Hallucination lie maker and publicly editable encyclopedia are not comparable. While it is true that you need to verify Wikipedia facts, you can assume it's a mostly trustworthy start. Nothing that comes from an LLM is trustworthy, every single thing it dumps needs to be verified w/ a *REAL* source
And just to not miss the thread here, the tweet I replied to said "Wikipedia" and "LLMs" are completely unrelated topics. You comparing how reliable they are as information sources means you agree with me that's BS, correct?
Thousands of humans working in concert or hallucinating misinformation.
Yep, identical.
![](data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="90" height="90"><rect fill-opacity="0"/></svg>)
As as longtime Wikipedian (and former employee in the Wikimedia Foundation's Communications team), I agree. A lot of the hate and bad press we got in the 2000s (and still sometimes more recently) was indeed similar.
The parallel don't just include hyperventilating about reliability, impact on students etc., but also e.g. "competes with copyright industry professionals and puts them out of their job" (we plausibly "killed" Brockhaus and various other paid reference works, by making knowledge accessible for free)
It’s quite interesting to hear that the analogy somewhat holds from folks who were around when Wikipedia was blowing up! Thanks for sharing that
I'm petty sure Wikipedia still is not a valid source by itself... If it has valid sources in its citations then those could be used.
so like when AI told people to make pizza with glue? or to get your daily intake of rocks? c'mon man
This is missing the point. This isn't just another, possibly-fallible-but-well-intentioned knowledge source. It's literally designed to be a plausibility engine. Its whole design is to generate plausible-looking outputs.
It is not built to deliver accuracy. It obscures wrong info by design.
Yeah, it's not a datastore. It's more for dynamic functions like summary, synthesis, styling, or translation
But OP is comparing the reaction to the 2 tools, and saying understand the tool that you're using
Wikipedia's sources are actual sources, AI's sources are sometimes sources and sometimes not and it has no way of telling which
Exactly. AI is also prone to sycophantry and is easily manipulated into changing its answer when challenged. You can get it to say just about anything.
I don’t understand what AI fanatics don’t get about this and it begins to feel super cultish to ignore the downsides and play dumb about it all.
wikipedia is put together by people who know information. AI is just throwing shit that kinda looks like information together. also, eat shit.
Wikipedia ISNT and SHOULDNT be a source. It is however, a curated collection of REAL sources that you can look at. As a starting point for a topic, Wikipedia is great at what it is.
LLMs are neither good as sources NOR good as starting places to explore sources. This comparison sucks.
More
Please throw yourself into a woodchipper if you think robots are so neat
Ai literally just repeats wikipedia at you and if it doesn't it just repeats jokes from reddit as if they're actual sources.
Ai doesn't care if its accurate, it just repeats things it's heard like a shitty game of telephone
Agree! I wrote about this last year!
open.substack.com/pub/mikekent...
The primary difference being that Wikipedia articles have sources, whereas AI engines don't.
AI fucking sucks ballz. It doubly sucks because corporations and policymakers are all in on AI mania, indicating that a machine will be our bosses and elected representatives soon enough.
I think we’re focusing too much on “AI is bad at stuff” and not enough on how genAI uses insane amounts of water and power to operate, that not only jeopardizes our already fragile environment, but also offloads the power costs onto facilities’ residential neighbors.
now who is to blame we are missing clean nuclear power sources and better managed waterways.
That's a fancy way of saying "it constantly spits out incorrect information and wastes your time because it's a bullshit machine that just slaps together words that look like they go together"
Tech Bros built a plagiarism engine to allow them to steal the work of artists and writers without ever compensating artists and writers.
it is extremely not a “piece of information” though
