how robust is rl training wrt collaborative agent behavior? does it come up naturally, or how much do you have to train for it?
collusion wiki could be a consequence of grader/env requirements and continual learning alone
link to the repo i used: github.com/brausepulver/clau…
