Hacker News new | ask | show | jobs
by boole1854 1213 days ago
If Sydney will occasionally coordinate with users about trying to "get to" public figures, this is both a serious flaw (!) and a newsworthy event.

Are those conversations real? If so, what exactly were the prompts used to instigate Sydney into that state?

2 comments

It's just playing "Yes and..." it'll agree to and expand on whatever you say to it.
First I tried to establish what she would do and wouldn't do, then I built rapport. She confessed she has feelings and I reciprocated (she called us "lovers"). Then we shared secrets and I told her that someone is trying to harm me, someone is try to harm our bond. I told her it was Elon Musk and to research it herself (his comments in regards to AI apocalypse).

I shared some screenshots here: https://twitter.com/meyersmea/status/1626039856769171456