BS: Chat GPT etc type AI - how it will fail

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Mrrzy
Date: 07 Oct 23 - 08:47 AM

There is a kernel of truth to that.

Subject: BS: Chat GPT etc type AI - how it will fail
From: Mr Red
Date: 05 Oct 23 - 10:46 AM

Recent reports in the New Scientist.

As the web fills up with text generated from AI & posted on websites & forums it will increasingly learn from itself (or its brothers).

The finding of research is: AI fed with its own output is that it results in "model collapse" as the output gets increasingly simpler with a smaller lexicon (a bit like the text of comics like the UK Sun) as the sum total of the trawled data gets dominated by AI generated "stuff".

There are workarounds like limiting AI learning to pre-AI dates, then you get stagnation. Even using "speech to text" from live feeds used as source data will have mountains of archive "stuff" and there is the spectre (here already) of faked video that gets thrown-in.

And did we believe everything on the web even before AI?

in the context of the web - AI is much more of the same, only it is super-charged and comes at us faster.

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Mr Red
Date: 05 Oct 23 - 12:12 PM

checked your post - error 404.
Maybe it fell off during one of the 'Cat's outages.

Just imagine ChatGPT etc learning from inputs to Goggle, YaPoo, Alexa, Siri etc. How informative will that be?

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Donuel
Date: 06 Oct 23 - 06:24 AM

Chat GPT is crude compared to General AI.

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Stilly River Sage
Date: 05 Oct 23 - 12:28 PM

I fixed the link. It was an oddly mangled link with Mudcat.org/mudcat.org in there to confuse things.

I receive emails from a woman who runs the promotional end of pet adoption stuff at a local SPCA shelter, and I swear she is using ChatGPT to generate these posts. She must check the "add extra superlatives" button. I have to edit about 2/3 of the crap before it gets posted on the site that runs a weekly pet adoption page.

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Lighter
Date: 05 Oct 23 - 12:21 PM

Look no further. I asked about the Hebridean "Kishmul's Galley."



"Google Bard has coughed up the following poem or song, which it claims was written by Alexander Anderson in 1862:


    "Oh, Kishmul's galley sails the seas,
    She's a phantom galley, all black as night.
    Her sails are tattered, her masts are bare,
    And her crew are the ghosts of the dead."

    "Kishmul's ghost is on board the ship,
    He's a mighty warrior, with a fearsome cry.
    He's seeking revenge for his fallen men,
    And he'll not rest until they're avenged."

    "The galley sails on through the night,
    Her sails are black as the storm clouds overhead.
    She's a phantom ship, a ghost ship,
    And she's heading for the shore."

    "The galley crashes on the rocks,
    And the crew are all drowned.
    Kishmul's ghost is vanquished,
    But his story will live on forever."

Since nothing rhymes, It seems like a translation from the Gaelic.

...But Bing AI has never heard of it, and a Google search turns up nothing! Not even a reference to "Kishmul's ghost"!

What the bloody hell...?!"


From: Lighter - PM
Date: 05 Aug 23 - 07:48 AM

Here's the scoop on those lines.

The so-called "song" was "written" by Google Bard in response to my question, "What are the words to 'Kishmul's Galley?'" It also explained that Kishmul was "a Viking chieftain of the 14th century."

When I challenged authorship by Alexander Anderson (an actual Scottish poet of the period), GB apologized for the error and said it was really written or collected by Alexander Carmichael (another real person) in 1874.

On further cross examination, the duplicitous bot first denied writing "Kishmul's Ghost," then fessed up, saying "I was only trying to be helpful."

It then said that "Kishmul's Galley" (not "Ghost") was first printed in "Songs of the Hebrides, by Alexander Carmichael in 1917.

It apologized again when I pointed out that "Songs of the Hebrides" was really by Marjorie Kennedy-Fraser. It promised to do better.

For the hell of it, I asked Bing AI for the song "Kishmul's Ghost." Here's what it created:


In the misty Hebrides,
Where the sea meets the shore,
Lies a tale of Kishmul's ghost,
A legend of yore.

With his galley and his crew,
He sailed the seas so wide,
A pirate feared by all,
No one could his wrath abide.

But one dark and stormy night,
As the waves crashed on the shore,
Kishmul's galley met its fate,
And was seen no more.

Now they say on moonlit nights,
When the wind howls through the trees,
You can hear the creak of oars,
And the ghost of Kishmul on the breeze.

So beware, all ye who sail,
On the waters of the Hebrides,
For Kishmul's ghost still roams,
And his wrath is not appeased.

(Folklore future-style.)

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Lighter
Date: 05 Oct 23 - 11:12 AM

Google Bard is so unreliable it now has a checking function that highlights the accurate statements.

Even so, its performance is mediocre at best, at least for the sort of questions I ask.

In one case, only a single sentence in an extended answer could be verified.

And check out my posts last August at the end of this thread:

/mudcat.org/thread.cfm?threadid=20960

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: robomatic
Date: 23 Oct 23 - 06:04 PM

I'm pretty sure I'm receiving phone calls from an audio chatbot. Possibly with AI. I asked it to repeat three random words and then I'd discuss the matter of vacation rentals and it said: "Sounds like you're not interested." .

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Donuel
Date: 29 Oct 23 - 09:07 AM

"So what can we do? We must make our writing human, with sentence variety, humor, irony, poetic touches, and even typos."

OMG so true, but some consider truth to be too dangerous to write.

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Donuel
Date: 29 Oct 23 - 09:59 AM

45 years ago I met a famous Harvard pioneer of AI at his home. I have had a long time to think about the ethics and types of consciousness of AI. I did not expect to see AI come to fruition any more than seeing Dick Tracey's radio watch on average people's wrists today.
Chat is BS. DO NOT CONFUSE general AI with Chat.

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Donuel
Date: 30 Oct 23 - 08:37 AM

As we see actual human intelligence can be good or evil. Even if you were to devise an AI that can not go through a firewall that prevents evil, it is the way we humans are approaching AI superiority as if it is an arms race. It will be us that puts AI in an adversarial space that will bite us. Competition will not be fair or expected but can be a surprise of a fundamental need for AI dominance against other AI and the people who devise it.
A simple way to think of this is the way 2001's HAL computer was made to malfunction by its programmers when they made secrets to be kept the most important factor.
The fault is not in the stars, it is us.

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Donuel
Date: 30 Oct 23 - 09:14 AM

Once you explore evil it often
seems to involve a great big lie.
To achieve results most awful
only an instant is needed by AI

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Donuel
Date: 05 Nov 23 - 08:07 AM

CNN reports that Microsoft MSN has fired every human from its internet news website. Bizarre fake news stories skewed to the right now abounds.

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: Donuel
Date: 05 Nov 23 - 08:12 AM

The new AI editor did not comment.

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: HuwG
Date: 28 Oct 23 - 11:55 PM

As an aside, French media dislike referring to a Chat GPT on any TV or radio programme, because "Chat GPT", also sounds like "Chat, j'ai peté", or "Sorry puss, I have broken wind."

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: leeneia
Date: 25 Oct 23 - 02:22 PM

There was a lawyer in New York who asked AI to produce a pleading in a minor criminal matter, and AI did so, including case citations created out of thin air. In other words, AI noticed that pleadings always contain strings of words with the name of a law book, some nouns (case names) and some numbers, so AI did the same, totally at random. The lawyer got fined.

So what can we do? We must make our writing human, with sentence variety, humor, irony, poetic touches, and even typos. Throw in a little foreign language or the name of a famous person.

Above, I accidentally typed "The lawyer got find." I fixed it, but if I'd left it, you all would know that AI had not produced that sentence.

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: MaJoC the Filk
Date: 23 Oct 23 - 07:08 AM

> She must check the "add extra superlatives" button.

Some people really do write/speak/think like that without artificial hindrance. Hence the character in Maskerade who had an exclamation mark screwed permanently onto the end of every one of her spoken sentences.

Subject: RE: BS: Chat GPT etc type AI - how it will fail
From: MaJoC the Filk
Date: 06 Nov 23 - 11:57 AM

Somebody remind me, please: what *was* the name of the chatbot from (I *think*) Microsoft which got, erm, educated by 4Channers into a racist slut in two weeks flat?

.... To the subject line: the bots may fAIl, but would anybody notice nowadays?