To be clear, I don’t think the complexity of intelligence matters for whether we should work on AI existential safety, and I don’t think it guarantees alignment by default.
I think it can confer longer timelines and/or slower takeoff, and both seem to reduce P(doom) but mostly by giving us more time to get our shit together/align AI.
I do think complexity of intelligence threatens Yudkowskian foom, but that’s not the only AI failure mode.
To be clear, I don’t think the complexity of intelligence matters for whether we should work on AI existential safety, and I don’t think it guarantees alignment by default.
I think it can confer longer timelines and/or slower takeoff, and both seem to reduce P(doom) but mostly by giving us more time to get our shit together/align AI.
I do think complexity of intelligence threatens Yudkowskian foom, but that’s not the only AI failure mode.