You've probably heard that AI chatbots can completely fabricate quotes. It happened yesterday in a DU thread.
The thread is in the Science Fiction forum, at https://www.democraticunderground.com/12002305 . I hope you'll read all of it. But see in particular the OP and replies 8, 11, 12, 15, 17, 19, 20 and 21.
DUer raccoon was trying to remember a partucular time travel story. There were a number of replies. One, from Goonch, had both a title and author, and a quote that was apparently the ending of the story:
Elias punched the coordinates for 1924 into the brass console, desperate to see his grandmother one last time. He pulled the lever, expecting the smell of ozone and the sight of her garden; instead, he was met with an absolute, crushing silence .
When the flash faded, there was no garden. There was no air. Through the reinforced glass of his pod, Elias stared at the cold, indifferent glow of distant nebulae [3]. He hadn't accounted for galactic drift—while he had traveled back a century in time, the Earth had continued its relentless orbit around the Sun, and the Sun had continued its 500,000-mile-per-hour sprint around the Milky Way.
He had reached the right time, but the Earth was billions of miles away . He was a ghost in the vacuum, a man who had forgotten that in the universe, you can never go back to the same place twice.
It looked like a pretty convincing answer. If I'd just run across it, I doubt I'd've thought twice about its accuracy.
But I knew it couldn't be right because I'd posted a reply hours earlier, mentioning the same story title and author, and linking to pages I'd found referring to it. I didn't have a quote from the story, and I've never read it.
But I had run across and linked to pages summarizing the story, about a girl named Marla (not a man named Elias) who went forward - not backward - in time, but also ended up in outer space.
So I asked about the source of the quotation above.
Turned out it had been fabricated by a chatbot. I've read lots of articles and social media posts about chatbots fabricating quotes, but it was the first time I'd seen this happen in a thread I was posting in. I thought it would be a perfect example to post here to show why chatbot responses should never be trusted without careful checking.
I want to thank Goonch for posting the explanation, which included the chatbot admitting the quote was apparently fabricated. Goonch also posted another helpful message with what the chatbot, Google's, had said later about chatbot dangers and the need to verify information, and it's an interesting read.
= new reply since forum marked as read
Claude's citation for the below text IS the thread that you and I are talking in right now
Verbatim paste:
When short fiction circulates online, people:
Change character names (Elias → Marla, etc.) to personalize or “improve” it for their own post.
Edit details like dates or the specific time‑travel instructions to fit a different prompt, meme, or context (e.g., Reddit writing prompts, AI demos).
Let AIs regenerate the same idea, which preserves some lines but changes surface details like names and wording.
Because there’s no canonical publication to anchor it, each repost or AI regeneration becomes “a version,” and they drift apart over time.
Is there an “original” version?
With no traceable first publication (e.g., an early dated story on a known author’s site, magazine, or book), there is no reliably verifiable “original” here. What you’re seeing is a meme‑like text: one early version got popular, and later copies, edits, and AI paraphrases created the Elias, Marla, and altered‑directions variants, none of which can be definitively proven to be the first.
In short: both versions are internet variants of the same micro‑fiction idea, not differing print editions of a book passage, and the differences come from casual edits and AI/reader rewrites rather than from an author‑sanctioned original.
In fact, I use Claude all the time, and it is very solid at getting quotes right when the attribution is clear-cut, but not when it finds a forum discussion thread like this. You also don't know the prompt that provoked this response which was very lackadaisically composed on my part.
AI getting this completely right would be outside the scope of a cheap or free subscription. This discussion, over multiple threads, involving multiple people, is simply too complicated for it to parse, given there's a limit to how hard it will work on any given response. All that it 'got completely wrong' is that it didn't search the entire history of the entire internet to see if the blurb was from any publication ever posted to the internet.
If I had a $100/month subscription using a model with more horsepower, and I wasn't using a model that specializes in coding work but rather on interpreting conversations, it would've done a better job. You get what you pay for in the AI world, and you need to use the right tool for the task, and you ABSOLUTELY have to compose good prompts 
But your overall point is taken: It's established that AI makes mistakes and should be reviewed.
P.S. I understand how and why "writing" generally, and proper attribution in particular is very near and dear to you and understand why this mistake (in Goonch's post) is particularly irksome 
for what you expect it should've said, otherwise it's "inane". Can you show how that blurb about Elias is very closely matched in *text content* with the denouement from the existing story, using "Marla" instead, and that it is available publicly online? If not, you are demanding that it do something that I did not ask it to do, and perhaps even something it could not have been expected to do even if I had.
And yes, absolutely, too much reliance of AI for coding can and often will result in cognitive debt. One thing that AI is exceptionally good at, however, is documenting its code, which is something that not all devs are that diligent at doing. A company can absolutely incur similar cognitive debt when people leave a company and the code was not well documented by the people leaving.
AI doing coding is 100% here to stay, so discussing the non-perfection in using AI to do it is a bit pointless at this juncture. There's probably nothing it's better at doing, and it relies on publicly available/non-copyrighted documents to do that work, so it's also among the more morally acceptable things to use it for.
And yes, it *will* cost a lot of people their jobs, including my own someday I'm quite sure. There's a difference for me in the topics of "what it's good at" versus "the harm to humanity that will probably come as a result of it replacing human workers"
when you don't know what my prompt actually was?
AI needs to be used judiciously in helping one write code. It is EXTREMELY useful in the hands of an experienced coder who really takes their time and writes good prompts. The fact that many people are going to "use it wrong" and expect AI to read their freaking mind, or don't know how to properly construct prompts, is arguably not a problem with AI itself, as such.
I would argue the biggest problem is the fuckhead AI companies paying off the pols so they can rush to market before there's ANY protective standards for humanity. If we just MANAGED the shit well, it could really be a "damn good thing".
One thing I hope you get from me is that I may seem to 'stick up' for AI, I'm mostly sticking up for the tech itself, which is pretty damn amazing, objectively, despite your dissent on that thought. But I am generally NOT on the side of these companies that made it, how they did it, nor how they've rolled it out.
If you see me cheering for the companies, please do call me out, sestra
I see a big difference between the likelihood of occupying the exact same spot at 2 times & a cosmological prohibition of doing so.
We agree on the extremely low probability, especially given our inability to traverse interstellar distances.
But, there is nothing in astrophysics that say we cannot do it if we had a ST style warp drive.
The theory does, in fact, prohibit the opposite, because it would require the creation of mass (which hasn't happened since shortly after the Big Band), with the same quantum properties. The former is not possible; the latter stupendous improbable. (Heisenberg and all that)
So, we're on the same page that it would be extraordinarily difficult to occupy the same space at more than one time, but there is nothing in the theory that absolutely prevents it.
I'm not a cosmologist, but I've been an avid student/obersver of that field and the math is not terribly different than that used in quantum chemistry, which I do know.
On board?
I do know enough to say that whether or not there's a 'law' in this regard depends upon how you define "place", because in one definition, "place" also includes a time coordinate. In another (and admittedly much more useful in day to day life) one uses a co-moving coordinate system (I think it's called iirc ) that accounts for the universe's expansion (I think is the basic idea?), such that the concept of "place" actually does continue to exist across time. And using the latter (and more useful) definition, yes, you are totally correct in your assertion. Fair?
It shows the dangers of using AI without verification or disclosure. Even if you disclose, you should check the source before posting. I've seen Gen AI completely misstate things.
If people use AI this way, they really need to disclose it. And really, people should be disclosing sources when they share "facts" that aren't based on their knowledge.
