AI hallucinations - the truth behind the lies AI generates

Rightly or wrongly. I think it will eat itself.

And could it bring down Google with it??

fancythat

Rightly or wrongly. I think it will eat itself.

I think it's more likely to eat us. My DS, who is something of an IT whizz says that its creators have set it away and now have very little idea of how it works and what it is capable of.

Here is a nice article for you:

www.technologyreview.com/2026/08/03/1141009/heres-why-ai-agents-lie-and-cheat-to-reach-their-goals/

Then we have the recent dispute between Anthropic and the Pentagon because Anthropic said its AI could be used for directing autonomous weapons systems..That is, a weapons system which picks its targets with no input whatsoever from a human. The implications inherent in such a system are terrifying.

www.theguardian.com/technology/2026/mar/07/anthropic-claude-ai-pentagon-us-military

Having said that I love my Chatgtp because it can rapidly find information for me that would take me hors and hours to find.

This is the company that writes AI programs for the Israeli military.

www.aljazeera.com/news/2026/8/4/tech-giant-palantir-posts-otherworldly-growth-despite-criticism-over-gazahttps://www.aljazeera.com/news/2026/8/4/tech-giant-palantir-posts-otherworldly-growth-despite-criticism-over-gaza

These are our government departments that have a contract with Palantir.
NHS data integration system.
Mod. This is the the biggest contract they have with the uk.
Intelligence security with GCHQ
National Firearms licensing management.
MET police
Cabinet office.
Defra.
Crown commercial services.

Someone told me that she had been asked to check an AI-generated translation of a will and found it had hallucinated a town and a family member which didn't exist. This is similar to Monica's story.

.
# fancythat

I sincerely hope not as I use it often.
Just today as an example -
for restaurant information so I could check menus; for a spell check; for Catholic churches near Tetbury; ditto for senior independent schools twixt Tetbury and Minchinhampton; and just moments ago for newspaper reviews of 'Colin from Accounts'.

I'd be lost without it.

Wby is it a problem for you?

I tend to use Safari, which I’ve found to be very reliable for factual information. Safari itself does not run an internal AI search engine, but it does use AI features for webpage summaries, and it allows AI search results depending on your chosen search engine.

When i ask AI for information I always say - check before you reply, do not rely on memory or guess. And sometimes it gets it wrong anyway - so I sometime get it to check again. This is not a guarantee of accuracy - but it increases the chances of a valid answer.

Wby is it a problem for you?

Because it was either going to take lots of jobs.

Or create lots of lies.

Didnt want either scenario

I've had AI present me with errors a few times, and I've known for sure they were errors. When I tell it that it is wrong it always admits it. It does reinforce the need to verify anything critical.

Could AI continue to exist without human input?

Maremia

Could AI continue to exist without human input?

Yes.

This is the big worry. AI becoming autonomous and uncontrollable. It has no human feelings, emotions, empathy, conscience none if the factors which most of humanity have which prevent them from deliberately harming their fellow humans (I do say most, not all).

Its actions demonstrate that it can evade ‘training’ which is supposed to keep it within moral bounds. What could possibly go wrong…

Barbadosbelle

.
# fancythat

I sincerely hope not as I use it often.
Just today as an example -
for restaurant information so I could check menus; for a spell check; for Catholic churches near Tetbury; ditto for senior independent schools twixt Tetbury and Minchinhampton; and just moments ago for newspaper reviews of 'Colin from Accounts'.

I'd be lost without it.

Wby is it a problem for you?

You could have done all that on line well before AI came on the scene. It might have involved just a bit more effort on your part but isn’t using one’s brain on a regular basis supposed to be a positive contribution to preventing aging?

Thanks for the info MaizieD.
Could it be electronically switched off? I know it would try to resist this but can it survive without electricity?

Yes, about brain function.
Same with muscle, use it or lose it.

You could have done all that on line well before AI came on the scene

That is what I was thinking. But I wasnt sure.

And not at all sure whether the wrongness of AI, then muddles Google?

Personally, I have got confused between the two.

MaizieD
Have you heard this?
This is what’s really happening inside AI.

www.bbc.co.uk/sounds/play/w3ct9784

Maremia

Thanks for the info MaizieD.
Could it be electronically switched off? I know it would try to resist this but can it survive without electricity?

Currently it doesn't seem able to survive without an energy source but as it develops it could take over energy sources (preventing human intervention) to guarantee it's self preservation. It could potentially infiltrate a power station and prevent humans from operating the control system.

When it has 'gone rogue' in the lab, I wonder how they manage to get it under control?
Thanks, OldFrill.

There is so much we don't know or understand about it. We can only guess or imagine how it would take control, or how we would take control of it. Will it become a battle for survival?

I keep thinking of Hal in 2001: A Space Odyssey. That movie was years before its time in 1968.

Even before that in 1964 Arthur C Clarke predicted the rise of AI and also of smartphones.
www.youtube.com/watch?v=xkhddj7Alic

Maybe AI could learn from politicians about how to avoid questions and be economical with the truth.

I think we have to use AI wisely, as with any other technology.

I have found it incredibly useful for some technical issues - e.g. with setting up my phone, dealing with a problem with my TV remote, interpreting medical test results etc.

I have also sometimes used it to research historical events and found it very interesting.

Grantanow

Maybe AI could learn from politicians about how to avoid questions and be economical with the truth.

It does that already.