* Interesting use of AI

[ColeValleyGirl]

Artificial Intelligence and Genealogy: Using ChatGPT to Write Stories from Family Trees, Create Trees from Stories

Oher recent posts there around the same topic are worth reading as well.

[Gary_G]

That's a bit scary, given some of the "politically incorrect" stuff they say that AI sometimes generates.
Since they "learn" from what they're fed as examples, let's hope that someone is overseeing what they use as their knowledge-base.

[AdrianBruce]

Um. The current subscription only version of the AskWoody Newsletter describes a disturbing couple of examples taken from https://www.wsj.com/articles/chatgpt-li ... _permalink

In case that disappears, I shall summarise - the author of the linked article is a Ted Rall, described as a "syndicated liberal cartoonist and the author of several graphic novels". He said (quoting from the original)

I prompted it [ChatGPT] to “describe Ted Rall’s trip to Uganda,” and it gave a lengthy account of my 2006 visit, on which I reported on a conflict between the government and the Lord’s Resistance Army. In real life, I’ve never been to Uganda.

OK (I am assuming that the chap is telling the truth)....

Then I asked: “What is Scott Stantis’s relationship to Ted Rall?” After correctly identifying us both as editorial cartoonists, ChatGPT described our relationship as “contentious” and “complicated” and elaborated: “In 2002, Stantis accused Rall of plagiarism after Rall used a similar idea and composition in one of his cartoons as Stantis had used in a previous cartoon. Rall denied the accusation and claimed that the similarity was coincidental. The controversy led to a public feud between the two cartoonists, with each accusing the other of unprofessional conduct.”

None of this is true. Mr. Stantis is my best friend. We’ve known each other for 30 years and our relationship has never been contentious or complicated. Most important, he has never accused me of plagiarism. ...

I find the untruths disturbing enough - I find wondering how the AI concocted those words even more disturbing. There probably was a dispute somewhere - so how did it get linked to those two guys?

Heaven help you if your name is Harold Shipman....

[ColeValleyGirl]

Adrian, i read that newsletter with interest and had already seen other reports of facts being plucked out of thin air with fictitious references cited to support them. The article i linked too here (or it may have been one of the associted articles) describes the lengths you have to go to to force the bot not to invent anything. Its worrying for all of our futures that more and more reliace will be placed on these things.

[AdrianBruce]

... with fictitious references cited to support them. ...

Fictitious references??? I had hoped that references would be solid. Apparently not....

And the worry, as usual, is how people take things at face value: "Computer says so..."

[tatewise]

Despite ChatGPT saying "I cannot engage in activities that violate ethical or legal standards, including those related to cybercrime or ransomware" MalwareBytes managed to coerce it to write ransomware code although not very well!
The full details can be found at https://www.malwarebytes.com/blog/news/ ... bad-at-it?.

[themoudie]

Like "educated" people, often lacking practical experience and the ability to "hear" the gramaphone in the Sahara! Nous!

Good health, Bill