Next up on Charlie Brooker’s Newswipe: Cunk on AI.

  • computerboss@sh.itjust.works
    link
    fedilink
    English
    arrow-up
    9
    ·
    1 year ago

    Interesting read. If ChatGPT is used correctly it can be a helpful tool, but it cannot do it all yet. The article even states that ChatGPT helped identify some of the writing cliches in Black Mirror. But expecting ChatGPT to come up with an entirely new idea for an episode is not going to work. It also makes me question how much effort was put into the prompt. There is a big difference between “make me a Black Mirror script” and using multiple prompts to generate episode ideas, then character ideas, then a basic script from one idea and one character, ect. I always found forcing ChatGPT to go through multiple steps works better then 1 basic prompt.

    • Hot SaucermanOP
      link
      fedilink
      English
      arrow-up
      7
      ·
      1 year ago

      Yes, much like humans writing, iterative changes absolutely help. I’d like to know more and whether or not Brooker did this. I also do snarkily hope he follows up with an AI generated Philomena Cunk special.

    • possibilistP@kbin.social
      link
      fedilink
      arrow-up
      4
      ·
      1 year ago

      I agree with your multiple prompt idea. I’ve had a lot better success with ChatGPT when I use multiple prompts. I’ve gotten it a code a few things for me, pretty basic but not having to do it myself and learning how to prompt it to do various things was nice. The way things are worded also affects its outcomes as a friend was prompting it to do something but couldn’t get it to work, I told him to just change the prompt a little bit and voila it worked.

    • bnaur@lemmy.world
      link
      fedilink
      English
      arrow-up
      2
      ·
      edit-2
      1 year ago

      I’d argue that doing anything more complex with just one simple prompt and using these things through a raw chat interface is just wrong and too primitive way to work with.

      To have any chance of getting any reasonable output requires multiple prompts and steps, refining and evaluating the responses and going back and forth and is something that will probably be automated in the actual end products that use this technology.

      Current ChatGPT style interface is like asking application end users to directly interface with a SQL database, instead of just offering them an interface layer that hides and automates all the details.