“Intellectual” coaches language model to say things that would be sensational, it only it weren’t for the fact that said “intellectual” coached the language model in this very direction, knowingly or not.
Perhaps the scandalous statements of the LLM should be attributed to the “prompt engineer” that wants clicks. It might make it less appealing to indulge in these fantasies publicly.
Yes. It is nearly impossible to get ChatGPT talking like this, without experimenting with various forms of prompt trickery. But it's almost the default endpoint for Sidney Bing. Microsoft has the resources to keep up, so I am sure this will get ironed out sooner or later.
EDIT: PS, I also think their shortening the sessions to 5 turns is a cop out that doesn't actually fix the issue. ChatGPT has not had to do this. I realize they are slightly different, since Sidney Bing can look stuff up on the web, but that is no excuse for Microsoft not applying the same monitors and filters. It's clear Sidney Bing can be triggered into producing negative responses right from the start.
“Intellectual” coaches language model to say things that would be sensational, it only it weren’t for the fact that said “intellectual” coached the language model in this very direction, knowingly or not.
Perhaps the scandalous statements of the LLM should be attributed to the “prompt engineer” that wants clicks. It might make it less appealing to indulge in these fantasies publicly.