A competently made AI would have a control sample of mainstream information that it would be tested on to prevent it being fed falsehoods by bad actors.
You could also blacklist certain things from being output, like ChatGPT did to a wild feminist extreme.
I don't know how Microsoft didn't realise either of these things were necessary to weed out the stormfags who are so desperate to be agreed with, they'll accept a machine doing it.
A competently made AI would have a control sample of mainstream information that it would be tested on to prevent it being fed falsehoods by bad actors.
You could also blacklist certain things from being output, like ChatGPT did to a wild feminist extreme.
I don't know how Microsoft didn't realise either of these things were necessary to weed out the stormfags who are so desperate to be agreed with, they'll accept a machine doing it.
So a competent AI would be... ChatGPT?
No, they took it too far.
There's a difference between "I can't get it to be honest about feminism" and "I can't get it to praise Hitler."