• a_beautiful_rhind@alien.topB
    link
    fedilink
    English
    arrow-up
    1
    ·
    2 years ago

    CAI/LaMDA and PI are trained more on convos than facts and QA. So they appear more “real” and personable.

    I don’t think we have an open model like that yet. Trained, not finetuned. Hence no new Blake Lemoines and a distinct feeling of “lack” when interacting.

    That’s my crackpot theory.

    • FullOf_Bad_Ideas@alien.topB
      link
      fedilink
      English
      arrow-up
      1
      ·
      2 years ago

      Why do you think it’s important to differentiate based on whether model is fine-tuned or is a base one? Do you have sources to confirm that LaMDA that was used by this guy wasn’t just a generalist model like base llama 65b finetuned on conversations? Basically Samantha LaMDA.