hgmodernism.bsky.social profile picture

H.G.Modernism

@hgmodernism.bsky.social

2913 Followers

30 Following

Hi, I'm Hendry bi-spectral whimsical nerdcore https://www.youtube.com/@HGModernism

  1. HOLY HECK YOU GUYS IT WORKED! It's always time for healthy teethā„¢! I... I don't know what to say... Thank you for your excellent participation!!!

    So you’re probably aware that YTers have a tool to ā€œrespondā€ to comments with ā€œAIā€. I’m curious whether this tool takes any information from MY actual comments. Reply to my YT post and I’ll reply with a scathing PG insult! Then I'll update if my reply recs change!

    www.youtube.com/post/Ugkx0Kj...

    Experiment Time!!! So you’re probably aware that YTers now have an internal tool to ā€œrespondā€ to comments with ā€œAIā€. Mine are all kind of self deprecating an...

    Post from HGModernism - YouTube

    Experiment Time!!! So you’re probably aware that YTers now have an internal tool to ā€œrespondā€ to comments with ā€œAIā€. Mine are all kind of self deprecating an...

    2
  2. This is solved using RLHF! Humans reviewers are given a set of possible responses and then rank them based on which are the best. And... surprise surprise... people like flattery. So those rankings are used to create a reward model and the LLM is trained to create the kinds of responses people like

    2
  3. Another example is tone, LLMs are trained on data from internet forums, for example, so you might also get responses like "ok loser, start with two cups of flour if you can bumble your way to your kitchen you fat piece of.... blah blah" which again, is correct but not the tone people want.

    2