LLMs can unmask pseudonymous users at scale with surprising accuracy
arstechnica.com
Pseudonymity has never been perfect for preserving privacy. Soon it may be pointless.
  • plateee@piefed.socialEnglish
    5·
    2 months ago

    I’d be shocked if they weren’t feeding in old data. Anything can be training data if you’re desperate enough, including early 2000’s Myspace pages scraped by the way back machine.

    • knightly the Sneptaur@pawb.social
      2·
      2 months ago

      The wayback machine is too lossy and would have missed most of my written corpus, I’m talking about finding me in the full Twitter firehose from prior to 2008 and accounting for the fact that my writing style shifts notably with new slang.