4 ms·

I wonder how they build up their training set? That couldnt have been that easy.

• mtgx 9y ago
  You're assuming it will be anywhere close to accurate. Knowing Facebook's previous attempts at AI, I wouldn't expect much. I wonder what the false positive rate will be. "Dave, you should stop thinking about killing yourself!" - "But...I wasn't thinking about killing myself?!" "Come on, Dave, deep down you know you want to. You can't fool me! I know everything about you." "- Well, now that I think about it, watching Facebook all day does make me feel quite depressed..."
  • dominotw 9y ago
    >Knowing Facebook's previous attempts at AI Curious what these were.
    • DashRattlesnake 9y ago
      The content of "places" feature looks like it's managed by a machine learning algorithm, and it appears to be a disaster (from a user perspective). It incorrectly merges and proposes bizarre categories all the time. There was an bad issue with a place I cared enough about to try to correct, so I ventured into the "community" they setup to build their training set to try to get it fixed. It was such a depressing place. Hundreds of people tricked by gamification into working for free for a for-profit company, only to be frustrated and have their efforts shit upon by its half-ass algorithm.
• supermdguy 9y ago
  From the article: > Facebook trained the AI by finding patterns in the words and imagery used in posts that have been manually reported for suicide risk in the past. It also looks for comments like “are you OK?” and “Do you need help?”
  • tomjen3 9y ago
    Oh great, the next time somebody needs to organize a move Facebook is going to flag them for suicide risk.