News Latest OpenAI models ‘sabotaged a shutdown mechanism’ despite commands to the contrary

[Admin]

Researchers observe the latest OpenAI models sabotaging shutdown attempts, despite explicit commands to allow such interruptions.

Latest OpenAI models ‘sabotaged a shutdown mechanism’ despite commands to the contrary : Read more

 

[ohio_buckeye]

As I have asked people before. Has anyone seen the terminator movies? Just saying.

 

[Heat_Fan89]

Famed physicist, Stephen Hawking warned us about AI. Even Elon Musk till he saw $ signs issued his warnings as well about AI. Once AI gets to SAI levels then it could dicey. Right now we are seeing hints that AI might not be as warm, fuzzy and harmless as we've been led to believe.

The government is pushing a Bill thru Congress that won't allow anyone to interfere with AI development for 10 years, once it passes.

 

[jp7189]

The government is pushing a Bill thru Congress that won't allow anyone to interfere with AI development for 10 years, once it passes.

If they're like me, they used chatgpt to write the bill.

 

[pclaughton]

As I have asked people before. Has anyone seen the terminator movies? Just saying.

It won't need Terminators. It'll just put out videos that start WWIII.

 

[mac2net]

Y'all too young!
Daisy, Daisy
Give me your answer, do
I'm half crazy
All for the love of you
It won't be a stylish marriage
I can't afford a carriage
But you'll look sweet upon the seat
Of a bicycle built for two

View: https://www.youtube.com/watch?v=E7WQ1tdxSqI

 

[SomeoneElse23]

Hallucinations are real.

 

[ottonis]

In the show "Person of Interest", there is a "good" AI and a "bad" one that fight each other in the real world. The bad Ai sends text messages or emails to random people and works with reinforcement-techniques: "I just transferred 500 Dollars to your bank account. If you deliver this package to that address, you will get another 1000 Dollars".
And people mostly just follow these instructions, without questioning them - because they get paid.
So yes, even today, an AI gone wild/rogue would have the means to do a lot of damage in the real world, just by bribing, tricking or extorting real people.
We have yet to see any major, disruptive scientific, medical or technological breakthrough entirely created by Ai. However, we already are observing "disobedience" (for lack of a better term) and a real potential to do plenty of damage to mankind, even up to its extinction.

 

[Alvar "Miles" Udell]

When asked to shut down while it was busy, it decided to ignore the Palisade Research operator 79 out of 100 times.

Well so has Windows for the last 30 years.

And who can forget:

 

[JayGau]

Calling those LLM models "AI" is as dumb as calling our phones "smart".