Wikipedia is too often treated monolithically when its accuracy is being assessed. It shouldn't be, as Wikipedia is a common technical and regulatory system for encyclopaedic articles, and the accuracy of any given article is a function both of the behaviour of the Wikipedia system and the article's own subject matter.

Like the Linux kernel, it's possible for an individual working alone to make a small incremental improvement to Wikipedia, and for the beneficial improvements to be detected and preserved by others, such that they preponderate. This is what Yochai Benkler calls peers-based commons production in Coase's Penguin.

Wikipedia's regulatory system has a policy and an enforcement mechanism.

Wikipedia's policy is called NPOV, Neutral Point of View. This may sound like a sincere attempt to be objective on matters of fact, but in principle means to be agnostic as between the truth and its alternative. I have no conclusive evidence for the following smear, but the reason for this agnosticism is that Wikipedia's creator, Jimbo Wales, is or was an "Objectivist", an adherent of an anti-authoritarian anti-collectivist pseudo-philosophy invented by novelist Ayn Rand, with whose works I am unfamiliar as to read them would see me stripped of my degree faster than throwing a brick through the Senate House window or watching rugby league.

Wales doesn't want an authoritarianism about what is true, so Wikipedia remains neutral as to "point of view". With minimal funding, of course, Wikipedia can't actually afford a truth-reckoning authority, but that is beside the point, because in practice, NPOV is pretty benign.

Wikipedia enforces its policies; there's a dispute-resolution procedure. I've used it. It sucks. Material can be forcibly removed from Wikipedia on grounds of NPOV-violation, falsehood or non-notability. The cost of contesting a strongly-fought dispute of this character is many hours of one's time. It's not worth it.

The relative costs and benefits of truth and falsehood in a Wikipedia article depend on the subject matter. The better our understanding of something, or the simpler it is, the lower the cost of maintaining an accurate Wikipedia article about it. (Simple and well-understood are not the same: quantum gravity is doubtless trivial; turbulence is complex but we know a huge amount about it; biology or economics on the other hand presently have neither characteristic.) There are few benefits to anyone of inaccurate Wikipedia articles about the physical sciences or mathematics.

Quite the opposite obtains in the case of history, politics, religion, linguistics, et c. Linguistics? I like to tell people that the point of historical linguistics is rape and pillage, and certainly in areas of the world where ethnic tensions are high, linguistic/onomastic arguments (e.g., your village's name is actually from the language of my ancestors, ergo ...) are used to inflame passions. It is easier to come up with a plausible historical linguistic argument than it is to plausibly refute one. People sufficiently motivated by contrarianism or a desire to fit in with the local racists can combine this fact with the general, non-domain-specific, tools of intellectual dishonesty are not going to be stopped by Wikipedia's wrist-slapping dispute resolution procedures.

This is why hard sciences and maths coverage on Wikipedia is very good, and some linguistic articles are an affront to the human capacity to perceive and reason about the external world. The good reputation of Wikipedia's physical science articles will attach to the lies of the ethnic cleansers, and vice-versa.