Recursion into madness

Security researcher lcamtuf ran an experiment on recursive AI output and wrote it up onlcamtuf’s thing. The setup: feed a model its own work over and over and see what breaks. For images, the answe…