I don't even think that matters much, right? Current LLMs already out-compete humans at many tasks. I think we're already past the threshold, at least in some regards. That is to say, I don't think there is a hard line because it depends on what your testing criteria are.