relatively few of the generally accepted laws of physics

"relatively few"? Name two.

I don't quite understand your confusion. An AGI is a computer program, and friendliness is a property of a computer program. Yes, these concepts allude to mental concepts on our maps, but these mental concepts are reducible to properties of the nonmental substrates that are our brains. In fact, the goal of FAI research is to find the reduction of friendliness to nonmental things.

You can reduce an AGI to the behavior of computer chips (or whatever fancy-schmancy substrate they end up running on), which are themselves just channels for the flow of electrons. Nothing mental there. Friendliness is a description of the utility function and decision theory of an AGI, both of which can be reduced to patterns of electrons on a computer chip.

It's all electrons floating around. We just talk about ridiculous abstract things like AGI and Friendliness because it makes the math tractable.

It does apply to every person. The other information you have about the claimed messiahs may allow you to conclude them not worthy of further consideration (or any consideration). The low prior makes that easy. But if you do consider other arguments for some reason, you have to take them into account. And some surface signals can be enough to give you grounds for in fact seeking/considering more data. Also, you are never justified in active arguing from ignorance: if you expend some effort on the arguing, you must consider the question sufficiently important, which must cause you to learn more about it if you believe yourself to be ignorant about potentially conclusion-changing detail.

See also: How Much Thought, Readiness Heuristics.

Bacon doesn't seem to have any special impact on anything

Man, I hope you don't mean that.

He believed that the scientific method he developed and popularized would improve the world in ways that were previously unimaginable. He was correct, and his life accelerated the progress of the scientific revolution.

The claim may be weaker than a claim to help with existential risk, but it still falls into your reference class more easily than a lot of messiahs do.

[It's extremely unlikely that] Eliezer or someone else who looks like a crackpot will actually save the world

This is ambiguous.

The most likely parse means: It's nearly certain that not one person in the class [*] will turn out to actually save the world.

This is extremely shaky.

Or, you could mean: take any one person from that class. That one person is extremely unlikely to actually save the world.

This is uncontroversial.

[*] the class of all the people who would seem like crackpots if you knew them when (according to them) they're working to save the world, but before they actually get to do it (or fail, or die first without the climax ever coming).

Of course, but the price of the Spectator's Argument is that you will be wrong every time someone does save the world.

How so? I'm not saying it's entirely impossible that Eliezer or someone else who looks like a crackpot will actually save the world, just that it's extremely unlikely.

Because you are making a binary decision based on that estimate:

Given how low the chance is, I'll pass.

With that rule, you will always make that decision, always predict that the unlikely will not happen, untii the bucket goes to the well once too often.

Let me put this the other way round: on what evidence would you take seriously someone's claim to be doing effective work against an existential threat? Of course, first there would have to be an existential threat, and I recall from the London meetup I was at that you don't think there are any, although that hasn't come up in this thread. I also recall you and ciphergoth going hammer-and-tongs over that for ages, but not whether you eventually updated from that position.

Eliezer has already placed a Go stone on that intersection, it turns out.

Do you have any candidates in mind, or some plausible scenario how the world might have been saved by a single person without achieving due prominence?

Ok, thinking how close we are to AGI is a prior I do not care to argue about, but don't you think AGI is a concern? What do you mean by a weak link?