Anthropic’s new AI model threatened to reveal engineer's affair to avoid being shut down

12 points · 8 comments · view on lemmy.world

8 Comments

CthuluVoIP@lemmy.world · 95 pts · 1y (1 reply)

*because that’s what the prompt they were testing was designed to elicit.

yup.

its so bs thad for som reason the peeps r treatin this as if its a new thing...

like - if i prompt my qwen to be bold, have a moral compass n take actions accordin to thad..

yea - itll tell peeps bout my affair.. if i ~~had~~ ~~one..~~

EDIT: dis entices me to do similar bs now... thad be funi >v<

That's just an ad.

Anthropic keeps pulling this bullshit line of advertising. LLMs will make up stories when you ask them to.

Good thing no one found out! /s

who@feddit.org · 9 pts · 1y

What is even the point of this research?

Seemingly to prove the people who have the skill to build an AI system are exactly the people you shouldn’t let run an AI system.