Joshua Pepper wrote:
Wikipedia is great for learning, but is not a primary source, and so should not be referenced by primary sources in science, in order to avoid mutual confirmation bias
Such a statement might be slightly dangerous. At least in Germany, a written thesis has to be prepared as part of the work for a academic degree, and it has to include a declaration that all used sources have been cited. If you used a tertiary source like wikipedia, you better cite it at the appropriate places, see for example the case of Annette Schavan. Also note that many original research math papers list specific "private communications" with specific other mathematicians in the bibliography among other references.
I like to cite wikipedia in my questions and answers, because this nails down the notion I'm talking about, and makes it clear that it is "well known". Also the provided links to other sources are often really valuable.
When it comes to learning something genuinely new, I found Stanford Encyclopedia of Philosophy orders of magnitude better (for the subjects that it covers). The same is also true for in a slightly different sense for nLab, see for example the explanation of the red herring principle:
The mathematical red herring principle is the principle that in mathematics, a “red herring” need not, in general, be either red or a herring.
Frequently, in fact, it is conversely true that all herrings are red herrings. This often leads to mathematicians speaking of “non-red herrings,” and sometimes even to a redefinition of “herring” to include both the red and non-red versions.
With respect to reliability, it's often hard to notice all the minor and major errors. When I tried to apply some of the information I learned from the wikipedia article on the Dedekind-MacNeille Completion, I was surprised to find that the information wasn't as accurate as it seemed to me when I read it without trying to apply it. I would have to check in detail how much of this misinformation is still present in that article today, but my guess is that most misinformation is still present (even if transformed slightly to make it less wrong.)