• 3 Posts
  • 45 Comments
Joined 3 years ago
cake
Cake day: June 15th, 2023

help-circle



  • My memory is really good actually, except for names and dates.

    When I learn something new about the world, I can cross-link it to other things I know and the structure of understanding reinforces the memories synergetically.

    But names are just arbitrary, they don’t mean anything. “That guy’s name is David” doesn’t reinforce, and isn’t reinforced by, any other facts about the world. I’ve just got to remember that lone, isolated factoid, and I’m rarely successful.





  • Sometimes PR people don’t write defensively, at least not entirely. Sometimes there’s an error, and the correct PR response is to acknowledge the error and communicate intent to rectify it for the future. Being totally defensive in the light of an actual error can do more damage than gracefully acknowledging it.

    LLMs are trained on data. They learn from actual human content. They’re usually pre-prompted to be agreeable, professional, and diplomatic. PR writing is probably a good chunk of the training data used to inform LLM word and phrasing choice.

    You’re essentially arguing that an impressionist painting “reads like AI” because you’ve seen a lot of AI images generated by models trained on impressionist paintings.