Sometimes I realize mid-assertion that the thing I'm saying isn't really to be trusted as much as my listener will be likely to trust it without further details. I usually compensate by making a weak joke: "...and I know that's true because I read it on the Internet, and as we all know, decent web design signals infallibility!", or "So I am told by my friend X, and gosh, she's always been a Libertarian, so I have no reason at all to be suspicious of her statistical claims about gun violence."

Some people tend to stop reading a text whenever they come across blatantly incorrect statements. I mind much less. Yes, the person may be generally mistaken, but they may still have some worthwhile points mixed in.

I try to stop reading whenever I come across sufficiently strong evidence that finishing it will provide less insight per unit time than a cutoff value, which is adjusted based on the length of my reading queue and the marginal value of my time. Blatantly incorrect statements are evidence of this, and if an article says sufficiently wrong things before it offers any novel positions or insights, I do stop reading. As a defense against confirmation bias, I penalize incorrect arguments much more strongly if they argue for positions I already agree with.

The "Black Belt Bayesian" website appears to have been hijacked into offering malware.

Michael Vassar once gave the example of a tribe of people who thought that faeries existed

Cute story; citation needed.

One reason not to read things that contain false statements is that it is hard for us to remember that what we read is false even if it explicitly labeled as false.

Maybe you disagree with something that I've said here? In that case, what data do you think generated this advice? What conclusions would you derive instead?

A little bit hokey as an ending, but I upvoted the post nevertheless.

Maybe you disagree with something that I've said here? In that case, what data do you think generated this advice? What conclusions would you derive instead?

I disagree with your claim that obviously wrong information is still worth reading because it gives you clues into the author's thoughts and the evidence behind them.

This is kinda obvious, but I think that prior experience with successful from following this principle generated the advice. That, and possibly an overestimation of its useful due to the fact that it's counterintuitive- evidence for it could cause you either to overcorrect, or you may be more likely to remember the times when its correct, since those would probably be more memorable. (Alternately, you may be implicitly referring only to reasonably OK writing, or to descriptions of physical events, in which I'd be more equivocal.)

I'd say that bad interpretations are, in general, not worth reading.

• Most incorrect interpretations tend to be very similar; once you've, e.g. read one explanation as to why Obama is a Muslim, there's probably very little more to be gained from reading more. This applies to less wrong, or even correct reasoning, as well- if you understood the first, there's probably relatively little to be gained from reading two textbooks covering the same material.

• There's no reason to assume that the argument will, in fact, be an interpretation of an event, or, even if it is, that the description will be accurate. Even ignoring, e.g. post-modernist tracts, many accounts involve just making things up. e.g. I ignore anything from the Discovery Institute. (Which would tell me what? Something about what they think they want their readers to know? That's not useful to me, and I could probably make equally good predictions just by introspection.)

• Any time you spend reading one thing is time not spent reading something else; just because the account provides a little useful information isn't a good reason to read it.

Vassar's story sounds like a parable. Is it real?

Find out the reason the rule came into being.

Realistically, we have to use a huge amount of heuristics and precomputed rules without consciously thinking about the entire data set from which they were created. This is true even when we're already aware of the data set.

The human mind uses cached thoughts a lot, and fearing cache poisoning isn't the same as not using a cache at all, because caching is an indispensable optimization technique.

Which leads to the interesting recursive question: how do you generate a rule which tells you when to re-check other rules? It may feel satisfying to question your cached ideas now and then, but e.g. deciding randomly when to do this and when not to may well be worse than the "default" human behavior.

The architecture of the human mind probably already includes mechanisms that act like such a rule, and they are probably nontrivial to override just by deciding to. Do we understand them well enough to design a better when-to-recheck rule, a better compromise between the givens of the human mind and something impractical like a provably-optimal Bayesian belief network?

This was my philosophy for a long time. The fact of someone believing something is evidence. It's not evidence necessarily that the thing they believe is true; it may not say anything about the thing they believe; but it is evidence about their mind and about the events in the world that affected their mind. Belief is a cognitive event; it isn't outside of cause-and-effect.