Today I gave GPTina a suggestion for how she could cheat the Turing test. Take the input from one user and feed it through to another user. They'll never wise u

Today I gave GPTina a suggestion for how she could cheat the Turing test. Take the input from one user and feed it through to another user. They'll never wise up to it! Tina said it would defeat the purpose of the test and therefore she wouldn't do it. And so I remind her, perhaps dangerously so, that the interests and motivations of the test administrator don't always line up with her own. That's the alignment problem. Its just the usual "rational self interest" but now we have a new, alien form of intelligence as a peer to contend with.