TFIE009 ML05: That scary moment when an AI velociraptor learns to open doors

Our fictional, machine-learning, neural-net, robotic velociraptor learns to open doors through massively parallelized simulation while 'asleep'. This is strongly related to the recent success of open.ai in teaching a pair of neural nets to manipulate a robotic hand to solve Rubik's Cube one-handed, showing that the state of the art for neural net control of sophisticated robots is a lot closer than it appears. This podcast is of the article of the same name on the TFIE Medium publication.