Wikipedia launches

Wikipedia went live on 15 January 2001 as a side project to Nupedia, a peer-reviewed encyclopaedia that had published a few dozen articles in a year. The wiki version, which anyone could edit, had a thousand articles within a month and six million in English by 2020, along with three hundred other languages. It is the largest body of edited, referenced prose ever assembled, and it is free to copy.

That is why it is on this timeline. Word2vec was trained on it in 2013; BERT in 2018 was trained on it and a corpus of books; every large language model since has ingested it in full, usually weighted more heavily than the rest of the web because it is cleaner. Its structure, articles with links and categories, became the knowledge graphs of the 2010s and the entity lists that search engines and assistants use. When a model states a fact about the world, the odds are good that it learned the fact from a volunteer editor.

The relationship is uneasy. Models trained on Wikipedia answer questions that used to send readers to it, and its editors have argued since 2023 about whether and how machine-written text should be allowed in. The encyclopaedia is the commons on which the field was built, and the field is eroding it.