I find Chat GPT so evasive. It gives these happy, positive non-replies, never addresses the real point of a question, won’t rule anything out, and then you find out it’s all bullshit. In other words, perfect politician material.

  • JackGreenEarth@lemm.ee
    link
    fedilink
    English
    arrow-up
    0
    ·
    17 days ago

    Yeah, but try an uncensored LLM like Mixtral 8x7B Instruct. With the right preprompt it can be quite helpful

    • exu@feditown.com
      link
      fedilink
      English
      arrow-up
      0
      ·
      17 days ago

      Even Mixtral is somewhat censored as they, like many others, used ChatGPT to generate question/answer pairs for the instruct finetuning.

      There are finetunes that remove this censoring again like dolphin.