We asked seven AI models to do a simple task. Instead, they defied instructions

Mirror: https://xcancel.com/dawnsongtweets/status/203945108300597700...

The research from UC Berkeley was submitted here: https://news.ycombinator.com/item?id=47608445