Programming thread

  • 🇵🇦 Nuestro primer dominio localizado está en español en kiwifarms.pa. Our first localized domain is on Spanish on kiwifarms.pa.
  • Want to keep track of this thread?
    Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading.
    Create account
the person writing this article is retarded
at first gpt3 at first was only given
Kod:
write me a 4chan greentext

> be me
> bottomless pit supervisor
then it generated everything up until "what" and ONLY THEN he added "to do" and then it made up the rest

iirc thats how the old gpt3 playground worked
why would it stop generating at ‘what’? Regardless I’m skeptical it was written by gpt at all, the model has improved significantly since then yet somehow this is the one time in history an LLM has been funny
 
why would it stop generating at ‘what’?
tokens output probably stopped idk
Regardless I’m skeptical it was written by gpt at all, the model has improved significantly since then yet somehow this is the one time in history an LLM has been funny
because current LLMs are prompted to hell and back ("You are ChatGPT 5.0. You act as an assistant chatbot....") GPT3 in playground mode was not prompted so it could be funny and racist
 
the person writing this article is retarded
at first gpt3 at first was only given
Kod:
write me a 4chan greentext

> be me
> bottomless pit supervisor
then it generated everything up until "what" and ONLY THEN he added "to do" and then it made up the rest

iirc thats how the old gpt3 playground worked
While true, that doesn't change that the meat of the joke (the guy becoming a regular pit supervisor, which leads pretty directly into the punchline) was provided by the prompter, not the model.
 
While true, that doesn't change that the meat of the joke (the guy becoming a regular pit supervisor, which leads pretty directly into the punchline) was provided by the prompter, not the model.
yeah but ">become a" pretty naturally leads to "regular pit supervisor"
and even then gpt3 was still relatively bad so it couldve said anything but just "bottomless"
 
AI kids these days never touched a model that didn't have crazy internal prompts to make it look like it's a person.
If you don't get some schizophrenic autocomplete after typing a single letter, it's not the real deal.
 
Gen Alphas don't know what they missed out on. In 2019 you could trivially get a model to write a story about Osama Bin Laden repenting and becoming a Catholic.

I'm working on getting emacs set up on my personal box again after 13 years. Going whole hog on Guile and using emacs as my environment for it. Why Guile? Why not I guess. Something to do, alongside all my other personal projects sitting at 75% completion for years. 🧩
 
Gen Alphas don't know what they missed out on. In 2019 you could trivially get a model to write a story about Osama Bin Laden repenting and becoming a Catholic.
I remember AlphaGo beating top players back in 2015, and then finding about MarI/O.
Back then I though neural nets were the coolest shit ever. Even though about going into data science. Probably biggest blunder form financial standpoint, that I have not done it.
Then there was this Harry Potter AI stories which were the funniest shit ever.

Nowadays though, I have severe case of AI fatigue.
 
I remember AlphaGo beating top players back in 2015, and then finding about MarI/O.
Back then I though neural nets were the coolest shit ever. Even though about going into data science. Probably biggest blunder form financial standpoint, that I have not done it.
Then there was this Harry Potter AI stories which were the funniest shit ever.

Nowadays though, I have severe case of AI fatigue.
Don't blame AI or MLP for the sins of LLMs :mad:
 
Wstecz
Top Na dole