Judge decides case based on AI-hallucinated case law

[Scribe]

https://www.cbc.ca/news/entertainment/ashley-macisaac-ai-accusation-9.7026786

 

[Dannyalcatraz]

https://www.cbc.ca/news/entertainment/ashley-macisaac-ai-accusation-9.7026786

“I’m a FIDDLER! With an “F”!”

Seriously though, I just the other day saw a guy who was talking about doing a search for Kon Knueppel, a NBA rookie. The AI assistant provided all kinds of info on the player, among which was the note that he was African-American.

(He’s a white guy from Wisconsin.)

 

[Jfdlsjfd]

On the plus side, 28% get it right. Given that ChatGPT entered the public awareness 3 years ago, and that honestly it's not intuitive, having one person in four already understanding it is not that bad. Sure 3-out-of-4 persons will think it either stores all the data intact to regurgitates it or have a little tiny human in the computer typing an answer (those 6% must be the one who ascribe intent to it, after all, the tiny human might be a mean tiny human) so it's still worth reiterating how it works, but I don't think we had 28% of people understanding how the Internet worked in 1995.

 

[trappedslider]

It’s a series of tubes right?

 

[Ryujin]

https://www.cbc.ca/news/entertainment/ashley-macisaac-ai-accusation-9.7026786

Not again. First he was accused of being anti LGBT, when he's been a staunch ally, and now AI takes a shot at him

 

[Dannyalcatraz]

In other, related news, tests of AI show them not obeying orders to stop in various ways, for different reasons.

AI's Big Red Button Doesn't Work, And The Reason Is Even More Troubling

"It's going to be really bad for humans."

www.yahoo.com

Killswitches are a vital part of most technologies. Undependable ones are virtually useless, and potentially dangerous.

 

[Jfdlsjfd]

Killswitches are a vital part of most technologies. Undependable ones are virtually useless, and potentially dangerous.

True...

This is exactly what it sounds like. A human operator's command to an AI should not be ignored by the AI

And that may be why big red buttons for AI shouldn't be handled by the AI itself, like most technologies in need of a kill swtich. If your nuclear reactor killswitch depends on everything being OK within the reactor to shut it down (for example, being powered by the functionning nuclear reactor), it sounds like a basic design flaw. An AI killswitch is "shut down the computer" not "ask the AI to shut itself down". Like basically any computer program. If windows hangs, it ignores the shutdown buttons, yet most people would know how to stop the system. The article focus on the bad without mentionning the conditions the blog post states for the bad thing to happen (developping actual superintelligence, allowing AI to self-replicate, giving it access to fully autonomous systems). Two of which are science-fiction and the third a basic engineering flaw. In term of reporting, that's on par with LUXEMBURG IS A THREAT TO HUMANITY (if Luxemburg had a stockpile of milloins of nukes and a sociopathic dictator bent on killing all humans as a supreme ruler). At this point, this is just the paperclip apocalypse thought experiment.

 

[Staffan]

It’s a series of tubes right?

Well, it's not a big dump truck, that's for sure.