Also, you are never justified in active arguing from ignorance: if you expend some effort on the arguing

Reference class of "people who claimed to be saving the world and X" has exactly the same number of successes as reference class of "people who claimed to be saving the world and not X", for every X.

It will be smaller, so you could argue that evidence against Eliezer will be weaker (0 successes in 1000 tries vs 0 successes in 1000000 tries), but every such X needs evidence by Occam's razor (or your favourite equivalent). Otherwise you can take X = "wrote Harry Potter fanfiction" to ignore pretty much all past failures.