No. of Recommendations: 3
OpenAI was testing the hacking capabilities of an AI model. To me, this sounds similar to gain-of-function virus testing. An obviously bad idea. Well, that choice of analogy makes it sounds a bit harsh : )
It could also be likened to testing a new drug for safety.
The distinction being that their new model was not specifically designed to be a black hat hacker, they were mainly trying to see if it already was. I guess they got their answer.
Of course, to extend the drug trial analogy, while testing the drug they found out that someone in the trial gave their drugs to a friend, who then died.
xkcd.com: xkcd: Human SubjectsJim
