4 ms·

> It is like "Q42 P31 Q5" for Wikidata - everyone understands it, ... I don't.

• danbruc 3y ago
  It's a semantic triple [1] - subject, predicate, object - made out of Wikidata IDs. Some common ones one will quickly memorize and start to recognize if one works with Wikidata. Q42 is Douglas Adams [2], P31 is instance of [3], Q5 is human [4]. [1] https://en.wikipedia.org/wiki/Semantic_triple https://en.wikipedia.org/wiki/Semantic_triple [2] https://www.wikidata.org/wiki/Q42 https://www.wikidata.org/wiki/Q42 [3] https://www.wikidata.org/wiki/Property:P31 https://www.wikidata.org/wiki/Property:P31 [4] https://www.wikidata.org/wiki/Q5 https://www.wikidata.org/wiki/Q5
  • lanstin 3y ago
    Fascinating. I had no idea people are still working on this sweetly optimistic project, like the architects writing a catalog of data taxonomies but for the unfenced territory of humanities knowledge. Sort of a modern Principia Mathematica with the Gödel of GPT4 waiting to blow it away. Still 42 for Douglas Adams manually assigned token is a nice touch.
    • brimwats 3y ago
      much of the AI projects would not really be possible without linked data, so it seems a bit silly to call it optimistic. the two are connected
      • tasogare 3y ago
        [dead]
    • vrandecic 3y ago
      I have a half hour YouTube video on that topic, maybe you'll find it interesting, but likely too long: https://www.youtube.com/watch?v=WqYBx2gB6vA https://www.youtube.com/watch?v=WqYBx2gB6vA
      • lanstin 3y ago
        It was interesting. I don’t think cheap matters much in a world:where compute is still getting rapidly cheaper. Being certain is good, but the throwmtons of data i to massive models keeps beating everything else. I do think the models need to get some self doubt built in some how but I don’t think the data catalog will be more than post NOC fact checking. Fun times tho.
        • lanstin 3y ago
          Edited: It was interesting. I don’t think cheap matters much in a world where compute is still getting rapidly cheaper. Being certain is good, but the throw tons of data into massive models keeps beating everything else. I do think the models need to get some self doubt built in some how but I don’t think the data catalog will be more than post hoc fact checking. Fun times tho. (My mom was an editor, she'd be horrified.)