I don't think the coherence part is particularly relevant here.
Consider two people, you (Peter) and me (Matt). Suppose I prefer to be able to murder people and you prefer that no one ever be murdered. Suppose I have the opportunity to murder someone (call him John) without getting caught or causing any other relevant positive or negative consequences (both under your preferences and mine). What should I do? Well, I should_Matt murder John. My preferences say "yay murder" and there are no downsides, so I should_Matt go ahead with it. But I should_Peter NOT murder John. Your preferences say "boo murder" and there are no other benefits to murdering John, so I should_Peter just leave John alone. But what should I do? Tell me what you mean by should and I'll tell you. Presumably you mean should_Peter or should_(most people), in which case, then I shouldn't murder.
(EY's theory would further add that I don't, in fact, value murder as an empirical claim - and that would be correct, but it isn't particularly relevant to the hypothetical. It may, however, be relevant to this method of testing metaethical theories, depending on how you intended to use it.)
EY-metaethics, as you have stated it, doesn't wash too well, since there is an implication that those who value murder should murder, those who value paperclips should maximise paperclips, etc.
Let me fix that sentence for you:
EY-metaethics, as you have stated it, doesn't wash too well, since there is an implication that those who value murder should_(those who value murder) murder, those who value paperclips should_(those who value paperclips) maximise paperclips, etc.
In other words, there is no "should," unless you define it to be a specific should_x. EY would define it as should_(human CEV) or something similar, and that's the "should" you should be running through the test.
Some will recognise that as a form of the well known and widely rejected theory of ethical egoism.
It isn't. Egoism says be selfish. There's no reason why someone can't have altruistic preferences, and in fact people do. (Unless that's not what you mean by egoism, but sure, this is egoism, but that's a misleading definition and the connotations don't apply).
But what should I do? Tell me what you mean by should and I'll tell you. Presumably you mean shouldPeter or should(most people), in which case, then I shouldn't murder.
There are a lot of candidates for what I could mean by "should" under which you shouldn't murder. Should-most-people woulld imply that.. It is an example of a non-Yudkovskian theory that doens't have the problem of the self-centered vesion of his theory. So is Kantian metathics: you should not murder because you would not wish muder to be Universal Law.
...EY-metaethics, as you
In You Provably Can't Trust Yourself, Eliezer tried to figured out why his audience didn't understand his meta-ethics sequence even after they had followed him through philosophy of language and quantum physics. Meta-ethics is my specialty, and I can't figure out what Eliezer's meta-ethical position is. And at least at this point, professionals like Robin Hanson and Toby Ord couldn't figure it out, either.
Part of the problem is that because Eliezer has gotten little value from professional philosophy, he writes about morality in a highly idiosyncratic way, using terms that would require reading hundreds of posts to understand. I might understand Eliezer's meta-ethics better if he would just cough up his positions on standard meta-ethical debates like cognitivism, motivation, the sources of normativity, moral epistemology, and so on. Nick Beckstead recently told me he thinks Eliezer's meta-ethical views are similar to those of Michael Smith, but I'm not seeing it.
If you think you can help me (and others) understand Eliezer's meta-ethical theory, please leave a comment!
Update: This comment by Richard Chappell made sense of Eliezer's meta-ethics for me.