It’s definitely not optimal. But our goal with these questions is to establish whether GPT-3 even has a consistent model of suffering. If it answers these questions randomly, it seems more likely to me that it does not have the ability to suffer than if it answered them very consistently.
It’s definitely not optimal. But our goal with these questions is to establish whether GPT-3 even has a consistent model of suffering. If it answers these questions randomly, it seems more likely to me that it does not have the ability to suffer than if it answered them very consistently.