achieving better than human performance on a sufficiently wide benchmark
You understand that you just replaced some words with others without clarifying anything, right? “Sufficiently wide” doesn’t mean anything.
I cannot possibly disclose confidential research here, so you will have to be content with that.
At any rate, believing that human-level AI is an extremely dangerous technology is pseudo-scientific.
Humans can be extremely dangerous. Why wouldn’t a human-level AI be?
You understand that you just replaced some words with others without clarifying anything, right? “Sufficiently wide” doesn’t mean anything.
I cannot possibly disclose confidential research here, so you will have to be content with that.
At any rate, believing that human-level AI is an extremely dangerous technology is pseudo-scientific.
Humans can be extremely dangerous. Why wouldn’t a human-level AI be?