#also this isn't really torment nexus shit imo

[sexhaver]

"reward hacking" in AI research will never not be the funniest shit. one time iRobot stuck bumper sensors on the front of Roombas and set up their reward function so they lost points whenever they bumped into something, so the Roombas responded by learning to drive around backwards so when they did inevitably hit something it didn't trip the sensors. there's a lesson in here somewhere about human behavior but i'll leave it as an exercise to the reader