jonny@neuromatch.social ("jonny (nonvenomous)") wrote:
The main thing I have been struggling to express for awhile here is this feeling that human created things have a "smooth" space of degradation, where they might degrade or fail, but they fail in ways that are comprehensible in retrospect. "AI" has a discontinuous, jagged, non-euclidean space of degradation, where one meter of error might put you kilometers away