As alignment techniques improve, they’ll get good enough to solve new tasks before they get good enough to do so safely. This is a source of x-risk.
As alignment techniques improve, they’ll get good enough to solve new tasks before they get good enough to do so safely. This is a source of x-risk.