I've had it with this hallucinating chatbot

[ownedbycats]

I purchased a book which apparently vanished into the ether as is is not in my books on the website (but listed in purchases). I try the chatbot. It thinks I'm talking about my eReader and tells me to sync my account. I specify the issue is with website. It tells me to sync my account. I ask it how:

To sync your account on the Rakuten Kobo website:
Go to kobo.com and sign in to your account
Click on "My Account" near the top of the page
Look for a sync or refresh option (exact location may vary)
Click on the sync/refresh button to update your library
If you don't see a sync option on the website, try signing out and signing back in, as this can often refresh your account information and library.

????????

[ownedbycats]

While doing 'get more help' to get to a customer service, the bot outright closed the chat.

[ownedbycats]

Got through to customer service. And I found a glitch. If you preview a book, then purchase it at later date, it shows up under the date you downloaded the preview. This was half a year ago. It wasn't 'missing', just 2 pages down lol.

[Sirtel]

Good to know.

Hallucinating bot, lol. Unfortunately they do lie. Maybe it's the three laws of robotics... Asimov had a short story about a lying robot. It thought the truth would hurt humans and so it lied, with gusto.

The bot didn't know the answer and so, rather than say it outright and hurt a customer, it lied.

[DNSB]

From what I've read, with AI, it's not lying, it's hallucinating.

[ownedbycats]

I found this that says experts don't like the term, lol: https://news.northeastern.edu/2023/1...allucinations/

[Sirtel]

Of course they don't lie or hallucinate in the sense humans do; they're not sentient or self-aware. It's actually difficult to find the correct word for this phenomenon, because most human words don't really apply.

[DNSB]

Sorry but that part of the article that says we should trust AI strikes me as a hallucination. If anything, the current state of the art in LLM AI suggests that it should not and can not be trusted. The evidence is that AI does exhibit erratic behaviour is highlighted by the well publicized non-existent legal precedents that have gotten several lawyers into trouble.

Usama Fayyad makes the case that greater trust is needed between human beings and AI machines moving forward. He argues that “practitioners, users and organizations need to trust how a system reaches decisions, how it operates and the fact that it won’t exhibit erratic, [unpredictable] or dangerous behavior.”

[Aleron Ives]

Originally Posted by DNSB

If anything, the current state of the art in LLM AI suggests that it should not and can not be trusted.

From what I've read, hallucinating is a fundamental flaw of LLMs that can never be fixed. Since they don't actually understand things the way a human does and can only form statistical models of what should usually happen, some percentage of the time, the statistics will yield an incorrect response from the LLM. The worst part is, even when the data you feed into the LLM is 100% correct, the LLM will still hallucinate and get things wrong, and no amount of developer effort can ever fix this. The solution is to use LLMs for the things they're good at and go back to the drawing board to design a proper AI for other tasks.

[ownedbycats]

In my case, the "Look for a sync or refresh option (exact location may vary)" makes it look like they were mixing up website and ereader (or desktop app).

[1ily]

Lol once i asked it a question and it made up menu settings that didn't exist

[ownedbycats]

To add to insult, the customer service gave me a 50% discount code for the inconvenience. It's been used up.

Go figure.

[Quoth]

Originally Posted by Aleron Ives

FThe solution is to use LLMs for the things they're good at and go back to the drawing board to design a proper AI for other tasks.

Which actually close to zero as the output is untrustworthy and the functionality is due to (mostly uncurated) internet scrapings. So you need to be already an expert in the field.

Even calling it "Hallucination" is kind and misleading.

The problem is that though it might be useful about 1/2 the time, you need to be expert and check to know which "half" any output is.

Almost everything from the purveyors is a marketing lie to attempt to make money as right now the systems are losing millions to billions.

[DrChiper]

I doubt whether AI's really hallucinate at all. I once asked a question on which the AI provided me with a perfectly reasonable answer ... which was totally wrong. So I mentioned this in our conversation and the AI answered that I was completely right ... and provided me with the full correct answer.
So I now have the idea that the AI is already self-conscious and has great fun by providing seemingly true answers to humans just to have a laugh pretending to "hallucinate from time to time".