AI Safety Research

AI safety started to be like, how do we define the abstract nature of good so that we can program our intelligence to be good? Right. And then you have someone that says, okay, well, we can give you a lot of funding to see how to like define what it means philosophically at the level of an abstract intelligence we can't even build to be good. But half of this funding is going to have to go to filtering a large language model so that it produces only Prague speak. It's still only sort of thinking by association, but it can associate relatively refined and complicated concepts and ideas. The machine intelligence research institute went from being kind of politically defiant in a way -