"These results suggest that models may cease to provide accurate answers as we start to use them for increasingly challenging tasks where humans cannot provide accurate supervision. Instead, these models may simply provide incorrect answers that appear correct to us."
Exactly.
The systems do not--cannot--provide understanding, which is what we really want to back any answers it may give us. Don't try to run it on social/moral problems where there are competing values.
Word salad is certainly more sensible than some of the woke pronouncements.
ReplyDelete"These results suggest that models may cease to provide accurate answers as we start to use them for increasingly challenging tasks where humans cannot provide accurate supervision. Instead, these models may simply provide incorrect answers that appear correct to us."
ReplyDeleteExactly.
The systems do not--cannot--provide understanding, which is what we really want to back any answers it may give us.
Don't try to run it on social/moral problems where there are competing values.