Even a pretty poor model (and it probably will be) would be a huge boon to editors in an encyclopedia full of references which are missing, out of date, a bit stretchy or plain bad faith. Wikipedia already has bots on it doing grunt work, and plenty of editors that like doing grunt work of checking minor details more than writing new content.
We're doing similar things to find errors in our regular ML datasets - a large proportion of the examples the model can't predict are mislabeled. Those mislabeled examples have a big penalty on performance. Since Wikipedia is often used in ML, it was time to clean it up.
Since it is open-source, I'd imagine some researchers with computer vision expertise and skilled with document/text digitization could greatly enhance this AI by digitizing these old books and including them in the training.
PDFs are a pain in the ass to read, especially on mobile
You can't use it to verify citations to books.
Edit: To clarify it could replace a citation to a book with a web page assuming that information exists on the internet.
Nothing that can't be done in principle, just hard/tedious engineering
I know, right? The halo effect is strong in this thread.
Ive got no love for FB or "fact-checking", but I knew much of this thread would be a dumpster fire.
(The video's context is the sinking of the German battleship Bismark. Whether or not you've any interest in that subject, the video can still give you a far better sense of the problem than Singularly Hub's article.)
Here is what the model is actually (attempting) to do:
>It calls attention to questionable citations, allowing human editors to evaluate the cases most likely to be flawed without having to sift through thousands of properly cited statements. If a citation seems irrelevant, our model will suggest a more applicable source, even pointing to the specific passage that supports the claim. Eventually, our goal is to build a platform to help Wikipedia editors systematically spot citation issues and quickly fix the citation or correct the content of the corresponding article at scale.