Well I think it’s a bit harder to parse out for most people. Effective altruism is easily seen as garbage when you learn the definition. AI Safety is a bit mor

Well I think it’s a bit harder to parse out for most people. Effective altruism is easily seen as garbage when you learn the definition. AI Safety is a bit more nuanced. I think there’s a lot of issues that are nonsense like having an “aligned AGI.” Aligned with whom? I could ostensibly envision you aligning it with a single person but aligning it with “humanity” is infeasible because humans are unaligned with each other. However there is an actual technical problem worth solving in alignment, which is basically matching the output of the AI with what the human intends not what the human asks. Moreover, AI has been improperly used already in ways that are plausibly avoidable. A classic example is algorithmically approving bank loans leading to members of ethnic groups getting denied at a higher rate due to their zip code, which is likely irrelevant to the loan. The designers of the algorithm weren’t explicitly trying to be racist (that would have massive legal ramifications on top of just being plainly unethical, but maybe they did it intentionally and I forgot), they screwed up and didn’t realize they produced an algorithm unaligned with their intentions. Figuring out how to avoid scenarios like that through either methodology or algorithmically is an interesting and much more tractable research problem than making an “AI that benefits all of humanity.”