AI blackmailing people

[Lookinagain]

I've seen a few article about this and it scares the you know what out of me. I've stayed away from AI for various reasons, but I know many use it. Does this disturb you?

Disturbing Signs of AI Threatening People Spark Concern

The world's most advanced AI models are exhibiting troubling new behaviors – lying, scheming, and even threatening their creators to achieve their goals.

www.sciencealert.com

Faced With Whether to Let an Exec Die in a Server Room, Leading AI Models Made a Wild Choice

All of the industry's leading AI models resorted to blackmailing users in new tests conducted by researchers from Anthropic.

futurism.com

 

[kenny]

I'm also VERY anti-AI.
I've never even been tempted to try it, even once.
I've survived almost 7 decades without AI so ... no thanks ... I'm good.

Once again I'm happy that I don't install apps on my iPhone.
Many apps are notorious for doing nefarious things without being noticed, and I expect AI will expand and exploit that exponentially.

I'm even avoiding updating the Operating Systems on my iMac and iPhone because they now include AI, or rather maybe more AI.

This way when AI problems (like ... OH ... blackmail ) show up, apple will hopefully build protections into their subsequent updates.
I'll hold out on updates till my old OS becomes a problem for me.

There are two kinds of users:
Those who camp out on the sidewalk in front of their Apple Store before the latest gizmo release, and set their device to automatically load every update when released.
Others wait for bugs to be worked out (by those apple-sycophant-suckers) before jumping into the water.
I'm the later.

Unfortunately this cat and mouse game will go on forever.

 

[Karl_K]

With the right prompts you can get them to say just about anything which is why the filters are somewhat jokingly said to be larger than the models.
Remember the data that went into these things is the internet, books, songs, and even movies and videos.
There is a lot of sick things in all of those sources.

So they take the output and put a muzzle aka filters on it so it doesn't say the wrong thing all the time.
Using the right prompts can get answers that get around the filters or a filter has not been added yet.