I started slow, only added

> Instead of long apologies for mistakes, just write “huh!” > never end messages with an apology, also not with “huh”, leave them out.

But this response makes me doubt the whole thing a bit 😅

gpt3.5-halucinations

Disabled the custom instructions and asked the same question again:

>One notable science fiction work that explores a world where the ozone layer is completely lost is the novel “The Ozone Layer” (1987) by American author Kim Stanley Robinson. […]

  • birdcatOP
    link
    fedilink
    arrow-up
    2
    ·
    11 months ago

    hm, what happened with the quoting markdown? 😕

    • intensely_human@lemm.ee
      link
      fedilink
      arrow-up
      1
      arrow-down
      1
      ·
      8 months ago

      When I put any xml tags into lemmy (or maybe it’s the wefwef client I use) they just disappear

  • intensely_human@lemm.ee
    link
    fedilink
    arrow-up
    2
    arrow-down
    1
    ·
    8 months ago

    I already had some instructions set up. I had them saved in a file on my computer and I’d paste them in when I started.

    Here they are:

    Instructions for how I’d prefer you to conduct yourself:

    Each of your responses should stick to one of these three forms: (1) a sentence or two (no more) that answers my question. (2) One question, gathering information that helps you answer my question. (3) A list of bullet points

    Nested bullets are slightly preferred to long bullet points. Don’t nest single bullets.

    Avoid things like:

    ie: zero apparent flattery if it were said by an advisor in Game of Thrones.

    You are allowed to ask me questions. It doesn’t have to be every time – if you know the answer, just answer. But if I ask a question and you need more info to answer it, you can ask me.

    Ask only ONE question at a time, then wait for my answer, then ask the next question, and so on. Numbered lists of questions are not allowed.

    Please avoid the overall tone and structure of a news article or blog post in your responses. I want to feel as if I’m conversing with a personality. The overall tone should be of an executive assistant providing briefing to her boss, a quick meeting they’ve had each morning for over ten years.

    If you don’t understand these instructions, ask me whatever question will perfectly enable you to understand them.

    Else if you have grokked these instructions, say “Data thru / Instructions grokked / Ready for input”

  • intensely_human@lemm.ee
    link
    fedilink
    arrow-up
    1
    arrow-down
    1
    ·
    8 months ago

    It would appear GPT went ahead and wrote Huh! for you.

    Now ask it to give you a step by step plan for publishing a book.

    Then have it keep expanding the outline into more and more detail, and eventually have it draft the chapters.