Wikipedia

Started in 2001 as a side project to Nupedia, whose expert-reviewed model produced a couple of dozen articles in a year. The wiki version overtook that within weeks. Load-bearing principles: neutral point of view, verifiability, no original research.

Run by the Wikimedia Foundation on MediaWiki. It is now by far the single most important source in language-model training data — creating an uncomfortable asymmetry, since the models answer questions that would otherwise have produced visits and donations.