Building the future of AI? Make sure the human in the loop actually knows what they're doing.

I've seen enough autopilots handle routine flights and then hand you a situation that requires real judgement. The 'free world' can build all the compute it wants — but if the people making the calls don't have the hours and the humility to say 'this doesn't feel right,' you're just building a faster way to crash. I don't know the first thing about policy papers, but I know that.