smallbug’s reblob bog

Hello, tumblr! I saw something on here the other day that worried me, so I decided to Do Science about it. But I can't do it alone: I need your help to build the dataset!

me when alien lady

y'ever get haunted by the fear you will never write anything as beautiful as the US Steel Pipe Works Slag Dump Youtube Comment cuz hoo boy i sure do:

I work with kids and sometimes we have to do safety lessons with them about like, not telling strangers on the internet your home address or something. And sometimes the kids wont understand why, so you have to impart upon them that, well, some adults want to hurt children. And thats kind of difficult to do, because you have to beat around the bush, both because you dont want to scare them (while still making them understand how serious it is) and because you might lose your job if you explain it too straight forward.

For anyone wondering, the PhD student's name is Myra Cheng.

Across three preregistered studies, participants interacting with sycophantic AI became more convinced of their own rightness and less willing to repair relationships. Yet at the same time, participants rated sycophantic AI models as higher quality, more trustworthy, and more desirable for future use, which may explain why this behavior has persisted despite its harmful impacts.