Resolves YES if there is an online encyclopedia other than Wikipedia on January 1, 2027 satisfying the following conditions:
- At least 5 billion words of original English, spanning roughly as wide a variety of topics as Wikipedia itself.
- The large majority (>90%) of text should credibly have been written by an AI
- Articles systematically meet the same or better standards of citation as Wikipedia, judged in terms of the broad veracity, extent, and variety of sources. I do not necessarily require that the competitor has licensing or particular citation policies (such as the No Original Research policy) matching those of Wikipedia.
- For a random sample of at least 100 articles on the competitor where there was an equivalent Wikipedia article in late 2024 (i.e. not recent events), I prefer the competitor's article at least 65% of the time over the late-2024 Wikipedia version of the article. I will use the following criteria in making a judgement: usefulness, accuracy, thoroughness, readability, and balance. Since I may be comparing recently-written articles on the competitor to 2024-era Wikipedia articles, which may be out of date, I will take great care not to let out-of-dateness be a factor in my judgement for the Wikipedia articles. I will factor article length in my decision-making to some extent, somewhat penalizing much shorter articles. However, I will prefer articles which use their length economically and effectively over exposition which is lower-quality or irrelevant.
- The project cannot be a "close modification" of Wikipedia itself. I will exercise judgement in determining what counts. As an example, if the competitor uses Wikipedia articles as a starting point for modification, or systematically heavily references the text of Wikipedia articles while writing, this would be disqualifying. It is fine if the competitor consults the titles and sources of Wikipedia articles (rather than the text of articles) in order to decide what to write about and/or to do research.
- It is fine if the competitor is not an encyclopedia in the standard sense, as long as its contents are systematically publicly accessible and indexable. For example, it's OK if the competitor's articles are primarily integrated into a search product, social media platform, or prediction market platform.
Update 2024-18-12 (PST): - A personalized AI wiki that is generated on-the-fly can qualify as a valid competitor, as long as it is publicly available and indexable and meets all other criteria in the description. (AI summary of creator comment)
@PaulHan Yes, in principle a wiki generated just for me could resolve this market as long as it’s publicly available and indexable. Inference would have to get really cheap for me to pay for 5B words of it, though. Not out of the question, but likely crowdsourced AI wikis will exist before personalized ones.