News for hackers. Stuff that matters.
Re:Science Is Shaped by Wikipedia: Evidence from a Randomize... by hliyan on Monday June 07, 2021 @07:59AM (#27421515)
Another idea for Wikipedia 2.0 is make the basic unit a "statement" rather than an "article". When Google Knol came out and called a Knol a unit of knowledge, I was excited that this was what they were going for, but sadly it was not to be. Now that we have GPT-3 like technology, it wouldn't be hard to generate generally readable prose from a database of statements. Each statement can be verified independently from each other, and can appear in multiple articles.

E.g.

  [Christopher Columbus]
    is a person
    is italian
    is an explorer, navigator
    birth 1451/08/25 - 1451/10/31
    death 1506/05/20
    is known for the exploration of the Americas
Re:Science Is Shaped by Wikipedia: Evidence from a Randomize... by smitop on Monday June 07, 2021 @08:06AM (#27421567)
That is the exact idea behind Abstract Wikipedia (https://meta.wikimedia.org/wiki/Abstract_Wikipedia), the idea of which is to automatically generate prose from structured data in Wikidata (https://www.wikidata.org). You can already do a lot of cool stuff with Wikidata today
Re:Science Is Shaped by Wikipedia: Evidence from a Randomize... by hliyan on Monday June 07, 2021 @08:24AM (#27421719)
That's great! The wikidata UI could be more compact though.
Re:Science Is Shaped by Wikipedia: Evidence from a Randomize... by cxr on Monday June 07, 2021 @09:33AM (#27422384)
See also the World Wide Graph <https://graph.global> by Mek from OpenLibrary/archive.org. It's more informal than Wikidata and has looser requirements. (Pretty sure Mek wants it to embody something closer to the original "sum of all human knowledge" spirit that was bandied about in relation to Wikipedia in its early days. Heed the privacy implications of this, however! <https://graph.5apps.com/LP/streamline/?id=660>)

I ran into the bugs in the default web-based client available from the graph.global homepage, though, and I'm not especially fond of the opinionated choices it makes about overriding browser-native behavior for dealing with navigation.

I made an attempt back in November at beginning to implement an alternative frontend for it. I have an "instance" (if you can call it that) running at <https://graph.5apps.com/LP/streamline/>, but right now you can only use it for read-only access to w2g. I will eventually fix this to allow graph changes, too, because my original motivation also included how cumbersome I noticed it was when entering new triples and linking together existing entities. I haven't worked out the right interaction for that yet. See this recent note <https://hypothes.is/a/RdwnFLRlEeu1mD-tdfIKCQ> about adding a command-line interface to w2g.

Granted, it has similar whitespace problems as your criticism of Wikidata. An even more streamlined version that is tailored to be accessed from the terminal is, I think, more in line with what you'd probably like to see.

Re:Science Is Shaped by Wikipedia: Evidence from a Randomize... by TomSwirly on Monday June 07, 2021 @09:01AM (#27422029)
> Now that we have GPT-3 like technology, it wouldn't be hard to generate generally readable prose from a database of statements

Which database of statements is this?

More, it might be easy to generate readable text, but accurate, correct text is a wholly different matter.