Caving Politics in the Nineteenth Century
- Thread starter langcliffe
- Start date
Did they dig under wharfedale to clash with Ribble?
www.ydrt.org.uk
DVRN Catchments -Yorkshire Dales Rivers Trust
Discover how the DVRN protects river ecosystems, improves water quality, and leads conservation efforts across Yorkshire.
www.ydrt.org.uk
langcliffe
Well-known member
As far as I know, https://halupedia.com/ is the only encyclopedia of halluicinations. Some of the articles are really quite surreal.
Well, there is already a dictionary, but:
An encyclopedia formed from AI hallucinations – what could go wrong?
Feedback discovers Halupedia, an online encyclopedia that is 100 per cent generated by AI, offering such delights as the 19nd century and The Society for the Prevention of Unnecessary Tuesdays
www.newscientist.com
langcliffe
Well-known member
Yes - that is where found it from.Well, there is already a dictionary, but:
An encyclopedia formed from AI hallucinations – what could go wrong?
Feedback discovers Halupedia, an online encyclopedia that is 100 per cent generated by AI, offering such delights as the 19nd century and The Society for the Prevention of Unnecessary Tuesdays
ZombieCake
Well-known member
What happens when AI starts ingesting its own hallucinogenic slop and adds it its pile of 'facts'? The downward spiral of what's left of humanity is well under way.
langcliffe
Well-known member
That is already happening, unfortunately.
semiretiredcaver
New member
langcliffe
Well-known member
This is what Anthropic's LLM, Claude, has to say about that:
"The immediate practical answer is: probably yes, eventually, for some models. Most large training datasets hoover up substantial portions of the publicly indexed web, and there is no obvious mechanism that would cause crawlers to skip halupedia.com specifically. Whether any particular model's training run includes it depends on when that run's data was collected, how the site is crawled, and whether it passes whatever quality filters are applied.
"The more interesting irony is the recursive problem this creates. If hallucinated "facts" get published on a website, then ingested into training data, they could in principle reinforce or spread those very errors — a kind of laundering of confabulation through the appearance of a citable source. This is already a recognised concern with the broader phenomenon of "model collapse" or data contamination, where synthetic or erroneous AI output pollutes future training runs.
"The counterargument is that the site's explicit framing as a hallucination encyclopedia might actually help, if future training pipelines become sophisticated enough to weight sources by reliability or flag known unreliable domains. Some current pipelines already attempt this. Whether the label "these are hallucinations" survives the chunking and embedding process intact is another matter — it very possibly doesn't.
"The deeper problem it illustrates is that the web is already substantially contaminated with AI-generated content of uncertain accuracy, and halupedia.com is really just a small, self-aware corner of a much larger phenomenon. The site is at least honest about what it contains, which puts it ahead of a great deal of what is already out there."
So the bottom line is, crap begats crap!
Cantclimbtom
Well-known member
No, it doesn't work that way
You can't really link us to what you read... because if I visit the page the content will have some similarities but be different. The content is belched up from the deep bowels of AI each time it's visited so will be different each time
Lankyman
Well-known member
I think a big problem in this age of mis-information appears to be many people's lack of critical thinking. It's been going on for many years, long before the rise of AI. I have a neighbour who confidently told me that COVID was spread by aircraft spraying it. When I challenged her over her conspiracy theory she got quite defensive (if you hear something along the lines of 'scientists don't know everything' then you know that person doesn't have a clue). People hoover this kind of twaddle up because it's easy and requires no more 'research' than viewing on a phone screen.
mrodoc
Well-known member
You have hit the nail on the head. It is what I have been saying for a while. It takes no time at all to check reliable sources on line (several if you are unsure). I am using AI increasingly but check the quoted sources as I go. It will extrapolate from data to come up with false answers sometime so if you ever doubt ask it where it got the info from.
Lankyman
Well-known member
Of course you don't necessarily have to check sources online there are (showing my age here) libraries full of print books. I had a library card from early childhood and used to devour all sorts of books on all sorts of topics. I can remember reading Eric Von Daniken at around 12 or 13 and thinking that sounds cool but a few years later realising that it was unfounded nonsense. An example often used by these types are the pyramids in Egypt - humans couldn't possibly have built them without alien help/technology. In reality the historical and archaeological record is quite clear on how it was achieved from simple mud brick mastaba tombs to increasingly ambitious structures over several centuries. And all freely available in many, many books written by credible authors based on real work. But that's harder than crap on the internet.
caving_fox
Active member
On the plus side, internet articles can get updated as new information becomes available. My parents have a wonderful set of printed books, a complete encyclopedia. It's missing the 2nd world war entirely. Some of the chemistry physics and biology is also wrong in ways that even online you wouldn't find today.
langcliffe
Well-known member
Are you sure? The content I have seen on repeated access to a page hasn't seemed to have had any obvious differences, and the home page of the website says: "Click any linked term inside an article to load its entry. New topics are documented at the moment of first access."
Lankyman
Well-known member
Yes, of course online content can be updated much more readily than printed. We rely on the good will and knowledge of the updater. I could read a book and realise it was outdated or inaccurate (I have lots of them at home including caving stuff). What I fear is happening is that a lot of people don't seem to be even aware that they're often being misled.
Similar threads
