Login
You're viewing the assemblag.es public feed.
  • Oct 15, 2026, 2:59 PM

    Goddammit, just realized the reason we need to identify stop signs and school buses and crosswalks to confirm we’re not robots is to train the AI in driverless cars.

    They’re cheating off our Turing test answers.

    💬 5🔄 15⭐ 5

Replies

  • Oct 15, 2026, 3:03 PM

    @batkaren I thought everyone knew that. See, this is why stating the obvious needs to be rewarded rather than mocked. Point and call prevents accidents. I have started saying "that's an ad" or "that's AI" to at least neutral reception, but "that's end-stage capitalist exploitation" is less well-heard.

    💬 0🔄 4⭐ 14
  • Oct 15, 2026, 3:06 PM

    @batkaren Part of this is that for a portion of reCAPTCHA, the system *does not know* the correct answers. It is possible to pass while still poisoning the training metadata.

    💬 0🔄 1⭐ 8
  • Oct 15, 2026, 8:31 PM

    @log @batkaren It's a mix of known valid/invalid images and ones that can't be reliably classified, so always pick the obvious ones and skip the dubious ones, then their machine learning learns the wrong thing while taking your response as reliable.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 1⭐ 5
  • 💬 0🔄 0⭐ 0
  • Oct 15, 2026, 5:19 PM

    @batkaren This is the reason for years I've done my utmost to fuck with them on the margins.

    "The presence of a bike lane, implies that the driver should be aware of bikes, so I'm going to mark that bike lane as a bike."

    "There's a bus stop there, which is part of the greater sociotechnical system that the bus operates in, so I see a bus!"

    "The full body of the rider is part of the motorcycle! It's not just the technical artifact!"

    This is why I have to go through like four captchas every time. Worth it.

    💬 0🔄 0⭐ 0
  • Oct 15, 2026, 5:47 PM

    @batkaren it's a good idea to answer those wrong a couple of times before you do it right, just to mess up their data

    💬 0🔄 0⭐ 0