[honoredb]

The story of a bug that caused the AI to optimize for maximally disturbing text that went unchecked because the only people authorized to stop it were asleep is a great illustration of how easy it is for an AI to "go evil" when you're not worrying about safety.

[GranPC]

Link, please?

[wgjordan]

https://openai.com/blog/fine-tuning-gpt-2/#bugscanoptimizefo...

(Someone didn't read the whole article..)

[ipsum2]

Not OP, but I think they're referring to: https://slatestarcodex.com/2018/10/30/sort-by-controversial/

Though I hesitate to draw strong conclusions from fiction.