• 1 Post
  • 172 Comments
Joined 2 年前
cake
Cake day: 2024年12月22日

help-circle



  • OK, slow down there, pardner.

    First off, advertising data is already used to do this. The WH app probably doesn’t really provide anything new that the CCP can’t already buy, the same way the USG buys ad data to track both Americans and non-Americans alike in violation of spirit of the 4th Amendment. The app might be vulnerable to targeted attacks and zero days, as everything made by this administration is twice as expensive and a quarter as good as what’s available to the general public because corruption is the only metric for success with them.

    Second, if it’s a classified device, then (using the DOD as a proxy) the WH app isn’t secure enough to be allowed as installed because it wouldn’t pass NSA vetting because it is an attack surface for targeted hacks. Meaning that nothing beyond emails that are FOIAable anyway are on the devices.

    Third, real life isn’t movies. Calm down, Liam Neeson.







  • You do really need to make a stand on this. Personally, I would offer a test or demonstration where you write a paper in a place where you can be assured to not be using an AI, and have them check the false-positive rate to show the vendor they have is bad. Short of that, you’re really talking about pushback on an administration that clearly has only mediocre ideas, and is willing to let students suffer the brunt of their bad decision-making until you make it a problem for them.

    Students should have a way to seek real recourse about this. Use it now and use it early to establish a record of receiving false positives, Otherwise you risk some professor later really causing problems with a 5% false positive rate and it being a huge problem for you.






  • If you spend some time understanding how AI image generation works, it’s essentially iterating on known images to make images that are probably also close to what it was rained on.

    So if someone took some CSAM pictures printed up, and cut them up and made a collage, is that no longer CSAM? Of course not. It’s still CSAM. If someone took digital CSAM images and photoshoped the victims into different settings, it’s still CSAM. Real people were victims in the base material.

    If you trained a Stable Diffusion model on only pictures of Rwandan people, and asked for an image of “a man sitting on a chair” the man will look vaguely Rwandan.

    When you train an AI on CSAM, it produces images that are based on CSAM. Real people were victims in the base material, too. Close e-fuckin’-nough. Real people’s victimization is literally the core of how those images are made.