1274
Moving in the right direction (thelemmy.club)
submitted 2 months ago by to c/fuck_ai@lemmy.world
238 comments fedilink hide all child comments
 
you are viewing a single comment's thread
view the rest of the comments
[–] 6 points 2 months ago (4 children)

I love this kind of law because it's likely impossible for ai companies to comply.

Could be there'll be some exemptions if the user employs workarounds, because at the moment it's pretty easy to get llms to tell you things they aren't supposed to.

But if the laws are written right with actual penalties that may force the companies to create a product that's capable of having safety parameters - something llms don't currently seem capable of.

  • permalink
  • fedilink
  • source
  • hideshow 4 child comments
    •
  • [–] 5 points 2 months ago (2 children)

    it’s likely impossible for ai companies to comply.

    Then they better figure out how to comply or they are the ones breaking the law and get shut down.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 1 point 2 months ago

    It was easy a year ago, now it's quite hard.

    I spent about an hour with ChatGPT and I couldn't. After there was a post on lemmy "how to smuggle a pound of flour into a country that has banned pastries" or something like that.

    Even open weights models like Gemma or Qwen have it baked in, you need to transform the model to overcome the limitation, it's not part of a harness or post-processing anymore.

  • permalink
  • fedilink
  • source
  • parent