We seem to be learning that we have a choice between safe, well behaved AI and creative capable AI.

I think the scariest thing for me about that is how poorly all the boomers in Congress will understand that.

I also though have some schadenfreude for big companies struggling with “we need to monitize this ASAP” and “we don’t want to get sued when our AI spouts Nazi propaganda and tells sick people to drink horse dewormer.”

I’m not sure what the news here is. “It’s easy to make machine learning models less safe on purpose” doesn’t seem surprising. The most concerning part is that you might make it less safe while trying to do benign fine tuning, but that info is buried in the middle of the article and not what it’s calling attention to.