• 0 Posts
  • 339 Comments
Joined 3 years ago
cake
Cake day: July 10th, 2023

help-circle
  • That would actually be the wrong thing to want. In an ideal system trust would always begin by the owner of the hardware, where possible, not the software or vendor they decide to trust.

    First the person that bought the system should take the ownership by overwriting the previous owners keys, and from there start signing the vendors key, they decide to put their trust in. Because it is important that the system is trustworthy to the end user/owner first.

    Any anti-cheat mechanism relies on not trusting the person that owns the hardware, and why would that be good?













  • In case of illegal numbers, intention matters. Because any number could be converted to different numbers, for instance through ‘xor encryption’ different ‘encoding’ or other mathematical operations, which would equally be illegal if used with the intention to copy copyright protected material.

    This was the case previously. You cannot simply reencode a video, a big number on your disk, with a different codec into another number in order to circumvent copyright.

    However, if big business now argues that copyright protected work encoded in neuronal network models is not violating copyright and generated work has no protection, then this previous rule isn’t true anymore. And we can strip copyright from everything using that ‘hack’.


  • I had a similar thought. If LLMs and image models do not violate copyright, they could be used to copyright-wash everything.

    Just train a model on source code of the company you work for or the copyright protected material you have access to, release that model publicly and then let a friend use it to reproduce the secret, copyright protected work.