This first bill allows the state of California to regulate and oversee all 3D prints in the name of public safety.

  • jballs@sh.itjust.works
    link
    fedilink
    English
    arrow-up
    0
    ·
    6 months ago

    Under the proposal, printers would have to evaluate STL files, CAD files, or other geometric code using a firearm blueprint detection algorithm and block files flagged as capable of producing a firearm or illegal firearm parts, including conversion devices.

    California’s Department of Justice, or another relevant state agency, would have until January 1, 2028, to publish performance standards for detection algorithms and software control processes.

    This is the problem when lawmakers write technical bills without speaking to technical people. They’re going to publish standards for evaluating if your gcode is a firearm or firearm part? THAT’S FUCKING IMPOSSIBLE

      • theoretiker@discuss.tchncs.de
        link
        fedilink
        English
        arrow-up
        0
        ·
        6 months ago

        Yes that’s probably how you would do this. Get a bunch of data of gcode of 3d printed gun parts and not-gun parts, for different slicers and printers. Then train some transformer as a classifier. Based on how good object recognition is, i would say its possible that you would get reasonably good accuracy and precision. And because you are scanning for code the architecture will likely be similar to an llm.

        • isleepinahammock@lemmy.blahaj.zone
          link
          fedilink
          English
          arrow-up
          0
          ·
          6 months ago

          Then five minutes later, someone figures out how to make a 3d printable gun that bypasses the gun detector on the 3d printer. It’s not like you’re printing a whole gun; you’re printing parts, most of which look nothing like a gun. How hard would it be to design an algorithm that takes a gun part cad file and then adds a bunch of extraneous pieces to it that can be easily removed? Just keep adding extra crap until the system no longer detects it as a gun part.

          • theoretiker@discuss.tchncs.de
            link
            fedilink
            English
            arrow-up
            0
            ·
            6 months ago

            Yes that would probably work. There could be some essential features of weapon parts that an algorithm might still be able to learn, and a printer could also keep track of previously printed parts for the classification. I think its unlikely that there are essential features of gun parts that are specific to gun parts so there would probably be a lot of false positives.

            • isleepinahammock@lemmy.blahaj.zone
              link
              fedilink
              English
              arrow-up
              1
              ·
              6 months ago

              Yeah I’m thinking of an automated version of greebling. Except you design the extraneous bits so that they’re only attached to the intended print like print supports - something in a non-critical location, easily torn away.

              • theoretiker@discuss.tchncs.de
                link
                fedilink
                English
                arrow-up
                1
                ·
                6 months ago

                I think a problem might be that the “easily broken away” might be a feature that an algorithm might be able to learn.

          • KelvarCherry [They/Them]@piefed.blahaj.zone
            link
            fedilink
            English
            arrow-up
            0
            ·
            6 months ago

            just rotate the piece at different angles on the plate, which would change the positions of vertexes and generate an unrecognizable set of printer head instructions. No extra pieces necessary; and even if there were, there’s no need for them to be printed attached to the main part.

            • theoretiker@discuss.tchncs.de
              link
              fedilink
              English
              arrow-up
              0
              ·
              6 months ago

              Isleepinahammocks idea would probably work. But rotation and translation would not. Thats something you can easily take care of in your training data, by reusing the same training data in multiple random positions and random angles.

                • theoretiker@discuss.tchncs.de
                  link
                  fedilink
                  English
                  arrow-up
                  1
                  ·
                  6 months ago

                  Yes i know. But this is trivially easy to get into the training data. You just slice models in random orientations and build your training dataset that way so the algortihm will recognize gun parts in every direction.