11 replies

1. iamdtms 2 min later

   @zachleat https://mas.to/@iamdtms/116511950365977966

2. cthos 🐱 33 min later

   @zachleat Why it isn't abundantly clear to people that the LLM makers are just frantically papering over the flaws that become memes? And then software devs are like "yep, I can definitely trust the text generator running in a loop to make quality software, boy howdy".

   1. Zach Leatherman 37 min later

      @cthos just one more model, one more

      1. cthos 🐱 40 min later

         @zachleat Bro, this last one was so much better than six months ago (repeat, until infinity, when nothing about the model has changed).

         Basically all that "advancement" happened in the deterministic tooling around the stochastic model - model didn't get that much better (and maybe got worse, considering all the goblin talk).

         I AM JUST SO TIRED.

   2. Alex Hyett 8 hr later

      @cthos @zachleat I am sure they have added "Strawberry has 3 R's" into the system prompt along with this recent addition: “never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user’s query.”

3. Paul Hebert 1 hr later

   @zachleat lol, rearmirrors would have been a solid guess for most "r"s if it was a real word

4. Lou Cyx 8 hr later

   @zachleat people actually let this things code for them 🙃

5. Sava 8 hr later

   @zachleat that's what you get when you start "learning" "advanced stuff" first and ommit the basics 😅

6. Eric Eggert 8 hr later

   @zachleat “AI” bro on LinkedIn was like “oh you are citing the strawberry stuff which has not been a problem in new models” and I just rolled my eyes because they don’t want to understand that there is no thinking behind it.

   1. Ollie Boermans 13 hr later

      @yatil @zachleat brrrrro

7. Zach Leatherman 18 hr later

   Gemini — are you okay