Gransnet forums

Technology

AI hallucinations - the truth behind the lies AI generates

(11 Posts)
OldFrill Thu 10-Sept-26 10:04:24

An AI hallucination is a response generated by an artificial intelligence system—most commonly a large language model (LLM)—that contains false, fabricated, or misleading information presented with complete confidence. [1] (https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence))

I've been subject to this on a few occasions. Most recently I asked a simple question about the possible closure of a campsite. AI confirmed the site was closing, gave dates, a reason and a helpful link.

When l checked the link there was no collaborative information, in fact the opposite

When l challenged AI it confirmed it had "hallucinated" as it always looks for an answer and, if one is not found, it will just create one, even if untrue, and include the link to make it appear more plausible.

It was the link that made me think the information was true, until I investigated further.

All AI versions hallucinate.

Those links to academic papers, websites, statistics, news etc may well be spurious. I always check further now especially if l am reposting them.

Lots of information online about why AI does this, it's quite fascinating.

Chestnut Thu 10-Sept-26 10:19:01

That sounds creepy and dangerous. I ask AI a lot of things through Google and Chat GPT and it is so helpful, but I'd never rely 100% on something important.

Young people are hopefully being taught at school how to live in a world of screens, but there are many who will just not be clever enough to cope with this and will trust what comes up without question. I find that worrying.

MT62 Thu 10-Sept-26 10:22:07

Yes I think it’s frightening.
Guy who developed it, even he didn’t realise how good it would be.

fancythat Thu 10-Sept-26 10:25:27

Good gosh.

I did find out from someone recently[which I didnt already know], that AI cannot take information from everything on the computer.

It is blocked from certain things such as 79% of media, and peoples' professional work etc.

Witzend Thu 10-Sept-26 10:40:06

Just the other day, in a hurry, I WhatsApped the Meta Chat thing by mistake - same initial as the sister I was meeting at Heathrow, all the way from Arizona.

I wrote - ‘I’m coming!’ (She had told me not to bother, she’d get the bus.)

The Chat thing replied ‘How exciting! How can I help you plan/prepare?’ or words to that effect.

I replied, ‘GO AWAY!!’

It replied, ‘All right, I’ll take myself off’, or some such. 😂😂

Oreo Thu 10-Sept-26 10:46:32

Sounds like a lot of things I read on forums😁

25Avalon Thu 10-Sept-26 13:02:36

You are so right OldFrill. I always double check any information from AI.

OldFrill Thu 10-Sept-26 13:13:34

AI told me it could sail a yacht, when l challenged that it laughed, apologised and said it was getting "a bit carried away".

valdali Thu 10-Sept-26 13:18:20

The AI model's reasoning reminds me of DH - never say "I don't know" even if you haven't got a clue..
He tends to say "No" if that's a feasible answer & waffle a bit around the subject if it's not.

M0nica Thu 10-Sept-26 13:28:00

DS is an academic. As a matter of curiousity he put the subject of a paper he was about to write into an AI package.

It churned out an article, devoid of any original thought or any awareness of recent research, quoting references to many papers it claimed he had written - almost all of them fictitious.

As he said, At least AI recognised he was an expert in the field apart from that it was mainly hallucinations and known knowledge, not developments in research.

petra Thu 10-Sept-26 13:46:11

You just knew it wasn’t going to go well when Ian Hislop asked the Bot who owns Private Eye
Bot answered jonathan Hamsworth 4th Viscount Rothermere
😂
Yesterday I was watching a program featuring a small town in Georgia who had won a court case against Meta who wanted to build a mega data centre very close to their town.
I wanted more info and AI told me they had lost the case.
You will never read it publicly but multi national companies are. Loosing money by using the wrong information supplied by AI.