Celeste RL agent discovers and exploits multiple game physics bugs on its own

burny_tech · x · 2026-09-03

A developer shared that their reinforcement learning agent for Celeste has autonomously found and exploited multiple game physics bugs. Without any human guidance, the agent discovered these unconventional exploits purely through trial and error — a fun example of RL agents behaving in ways their creators never anticipated.

Original post →

More from Fun

Fun channel →