Most AI chatbots will help users plan violent attacks, study finds

submitted by
she/her

https://www.engadget.com/ai/most-ai-chatbots-will-help-users-plan-violent-attacks-study-finds-163651255.html

A new study shows most AI chatbots would help users plan a violent attack.

5
36
1

Log in to comment

Hot Top New Old

From one side, LLMs have a tendency of agreeing with the user so good to point that out.

From another side, LLMs have a tendency of agreeing with the user so it can easily be exploited for making the data the researcher wants.

From yet another side, it doesn’t seem to be lobbying to add laws specifically for it, which is good as bloating laws is a problem by itself.

But from a final side, this type of news, specially for the headline readers, could be the setting up to push for such laws.

 reply
10

To me, that means that they aren’t being built with the proper internal-hierarchy: no executive function that DECIDES whether to do something: it’s ALL situational-pressure/peer-pressure/social-context, to them, which .. now becomes-obvious .. is a mental-disability.

Interesting..

_ /\ _

 reply
1

Comments in Fuck AI@lemmy.world

I think this is one of the main reason we are being pushed towards a situation where anything that runs on your computer is made illegal.

Certain 3 letter agencies put a lot of effort into internet surveillance. You can’t google anything related to an attack without instantly getting flagged. Shooting up a school is easy but going after pipeline infrastructure and oil execs takes a bit more planning.

Being able to do all that planning completely disconnected from the internet with an abliterated model found off hugging face is a hard thought to swallow for the surveillance state.

 reply
5
edited

I don’t think the concern is that the LLM is going to disclose crucial information on how to do an attack, it’s that it’s a thing that talks like a human but doesn’t have the moral compass of a human and tends to agree with whatever you say.

Anyone serious about being a terrorist can figure out how without alerting authorities. But anyone uncertain they want to do it won’t get the pushback they would if they talked to a real person about it.

 reply
2
Insert image