Thou shalt not create a machine to counterfeit a human mind.

  • 0 Posts
  • 16 Comments
Joined 8 months ago
cake
Cake day: December 14th, 2024

help-circle
  • As far as I can tell, the NSA data was into a dataset that allowed report software to run against it. It was also largely metadata, and it didn’t assign a person to the metadata.

    Meaning it wasn’t an “enter a name” or "enter social security number.

    This sounds like a dataset built for each person. Now how that’s going to work is a different question. Cops can already pull you over, and once they have your license plate, they can see if you’ve got warrants or outstanding fines, and various legal history.

    Palantir’s data sounds like an efficient way to cause mass amounts of identity theft.




  • So… I actually proposed a use case for NLP and LLMs in 2017. I don’t actually know if it was used.

    But the usecase was generating large sets of fake data that looked real enough for performance testing enterprise sized data transformations. That way we could skip a large portion of the risk associated with using actual customer data. We wouldn’t have to generate the data beforehand, we could validate logic with it, and we could just plop it in the replica non-prodiction environment.

    At the time we didn’t have any LLMs. So it didn’t go anywhere. But it’s always funny when I see all this “LLMs can do x” because I always think about how my proposal was to use it… For fake data.