Vechev and his team found that the large language models that power advanced chatbots can accurately infer an alarming amount of personal information about users—including their race, location, occupation, and more—from conversations that appear innocuous.

  • GenderNeutralBro@lemmy.sdf.org
    link
    fedilink
    English
    arrow-up
    5
    ·
    1 year ago

    LOL. Nice!

    I wouldn’t expect ChatGPT to be well-versed in forensic linguistics; I suspect a human expert could make better guesses based on seemingly-innocuous things like sentence structure and word choices. I’ve seen some research on estimating age and gender based on writing. There’s a primitive example of that here: https://www.hackerfactor.com/GenderGuesser.php

    My last comment is a bit short (it wants 300 words or more), but I am amused by the results:

    Genre: Informal
      Female = 338
      Male   = 309
      Difference = -29; 47.75%
      Verdict: Weak FEMALE
    

    I’ll pat myself on the back for writing more or less down the middle. :)