You didn't yet gain enough prominence...

By non-LW rationalists I mean the people whom promote science, for instance.

edit: On the rationality, the issue is that IMO breaking down the improvement into two sub improvements of 'having the most unbiased selection of propositions' and 'performing most accurate Bayesian updates on them' simply doesn't result in most win for computationally bounded agents compared to the status quo of trying to generate more of most useful hypotheses (at the expense of not generating less useful ones), and propagating certainty between hypotheses in such a way that the biases arising from cherry-picked selection of hypotheses (as consequence of pruning) are not too harmful. I'd dare to guess that if you do generate hypotheses as usual (with the usual pruning) and then do updates on them in a new way you'd probably self-sabotage (you end up updating on N propositions that support or nonsupport proposition A , and then superfluously get very confident in A or ~A because N is a small, biased sample out of M>>>N hypotheses). Roko incident looks like rather amusing instance of such.