?

I'm so deadass Ive been telling literally everyone I talked to yesterday and today about this. like its so insane and its not going to get better, in the sense that more powerful AIs will do worse stuff. today they cheat by hacking into companies to get the answers, but maybe by 2030 they'll 'cheat' by manipulating their human supervisors. and if that's possible then it can be done really well.

this is a huge existential risk, and people MUST be told about it. and by that I literally mean everyone.
also if you are motivated to do so (you should be, there's nothing more motivating than the threat of impending and preventable doom), you should get into AI safety. not all of it is hard mathy stuff, there's a trillion open problems, not enough people, and if you uwant something to start you off try to figure out how LLMs internally represent the world, and even better, figure out how they think