The state-of-the-art deep reinforcement learning methods take hours or days to learn how to beat even the first level of Super Mario Bros., but the best solutions for beating Super Mario Bros. that human can find consist of mostly a few actions. With this knowledge we hand made a biased random Agent that beats the first level in 0.5% of the time. Random input beats the level in fewer steps than the state of the art RL algorithm, but once the RL agent is trained it is a lot more consistent.