• python@lemmy.world
    link
    fedilink
    arrow-up
    2
    ·
    5 hours ago

    What kind of hyper-social person would even have enough text messages to train a model? Not only do I send messages like twice a week at max, my tone is completely different depending on who I talk to. Not even my language choice is consistent - when my parents text me in russian, I reply in russian (via bad speech to text, I can barely write that language), but when my husbands parents text me in russian, I usually reply in german. Coworkers and friends usually get german, unless I’m quoting english memes or technical documentation and my husband and I text in 95% english unless german is totally necessary. Just explaining that constellation to an LLM would be super annoying haha

    • Test@lemmy.zip
      link
      fedilink
      arrow-up
      1
      ·
      4 hours ago

      Most of my communication is text based on different platforms from good old rcs text message but mostly through online or app based platforms / services. I don’t think you would need enough to train a model but just to prompt as in the model already understands language and how to write and you’re just setting a parameter on what style to write much like how voice or video synthesis can take just 30 seconds of audio or just a few images of a person.