• 0 Posts
  • 39 Comments
Joined 7 months ago
cake
Cake day: December 29th, 2023

help-circle
  • Generative AI doesn’t get any training in use. The explosion in public AI offerings falls into three categories:

    1. Saves the company labor by replacing support staff
    2. Used to entice users by offering features competitors lack (or as catch-up after competitors have added it for this reason)
    3. Because AI is the current hot thing that gets investors excited

    To make a good model you need two things:

    1. Clean data that is tagged in a way that allows you to grade model performance
    2. Lots of it

    User data might meet need 2, but it fails at need 1. Running random data through neural networks to make it more exploitable (more accurate interest extraction, etc) makes sense, but training on that data doesn’t.

    This is clearly demonstrated by Google’s search AI, which learned lots of useful info from Reddit but also learned absurd lies with the same weight. Not just overtuned-for-confidence lies, straight up glue-the-cheese-on lies.




  • The lack of pressure leads to absurd file sizes for silly things.

    A few weeks ago, I needed a vector company logo, so I asked our graphics team for one. The file they sent me was 6MB. While working with it, I noticed it was actually quite clean, so I exported it as an SVG and it came out to 2KB. 1/3000th the size for the exact same graphic.

    I opened their file up in a text editor and found font configs for specific printer models (in a graphic with only filled curves), conditional logic, multiple thumbnails, and other junk.