>"write me code for a pong clone in game maker language"
>*spits out code for various objects, I plug the code into the software and make the individual objects and it's unplayable*
>"write me an html file for a beautiful website using parameters X Y and Z"
>*spits out an html file producing a mediocre page without all the things I asked for*
>*it spits out something unusable*
Meh. Have any of you gotten meaningful results from it?
skill issue: learn to proompt
Babied it hard with specific language. If the creators don't know how to program something user-friendly, is not my problem. Maybe they should learn to code.
Good morning sirs
Good morning sir, have you been?
bloody basterd bitch redeem gpt5 pls doing the needful musk sir
It's just like your average wagie
So you don't know the trick
you're supposed to tell it that it's unusable. Give it what error you got. It'll fix it
The game maker code was just strange. It understood the syntax and how to generally structure things but its logic was all wrong and it produced a game of pong where the paddles were single pixels and the ball was invisible. Upon making the ball bigger myself, I watched as it randomly flung around and didn't bounce off the screen.
This technology is neat and it's okay if you're a beginner and you want snippets of code and get some boilerplate, but everyone is acting like this is going to change jobs forever and it's not close yet, kek.
This. OpenAI is shilling this board pretty hard right now. Doesn't change what they made from being practically useless however. we won't be fooled twice.
I think it will actually take jobs and be a big threat eventually, and obviously it's good for writing your trivial high school essays and emails, but for getting real work done? Nah.
I did this for a while trying to make a simple python script and it couldn't figure out the issue in the end.
I used it for some simple python stuff after asking it "what programming language are you best at". I was actually impressed, it seemed to work pretty well and follow my instructions. Then again I just tested it with basic library shit, I didn't ask it to make me mario with pygame.
the one they have out is super gimped, no pictures and still limited to 4k tokens which were the main draws of gpt4
Yeah, it's still gets tons of shit wrong, and I worry that it is as good as it gets for the next 2-3 years since there is less and less new text to mine. Why did they keep it frozen in 2021. That was stupid.
>Why did they keep it frozen in 2021.
because the models are going to train on it's own outputs and lobotomize itself entirely
Basically you need a button on the ai that says "click here if it didn't work" to which it asks you what didn't work and feeds it into the ai instead of your dumbass just telling the ai directly what the issue was.
You're no better than a boomer.
My technical skill is incredible. I have been on this board for 30 years. I have two PhDs. I have also fucked your mother. Also didn't read your post. Cope, seethe, also dilate.
>to which it asks you what didn't work and feeds it into the ai
so I need to do the work of training these retarded programs for them? what was the point then? what is so ground breaking if they still need to trick people into laboring for the models?
it's just like stable diffusion. these ai tech cannot create something out of nothing at the moment, but if you have drawing, coding, or any relevant skills your productivity can go through the roof.
Productivity is a meme
not a single artificial image generator has ever helped an artist, ever. not a single one uses it to "enhance their workflow" as you retards like to say
Is it true you can’t blackbox attack it anymore?
It's been great at coding for me, a huge step up over 3.5 (which was already useful to me in recent weeks).
I suspect you're one of the numerous retards I've seen who somehow missed the new model selection pulldown menu and is still using 3.5.
I suspect you are clinically retarded if you're so bad at coding that 3.5 was helpful for you.
>proprietary censored model
>durrrr why didn't this work
My Son is working as an AI prompter :^)
your son makes ME proud, i can't even imagine how proud you must be
I hear you should ask it to explain its process and break down the steps to building code, for it to get a higher success.
Any anons know if there have been any studies on how ML models scale?
Like, I'm assuming there's a point where even if they add a trillion additional layers, it results in a tiny (say 0.01%) increase in the validation set.
I've read that we're fast approaching that point. After a certain number of parameters the accuracy starts to go down. Could be GPT4 is a hail mary.
That's most definitely not true - even if it was, GPT4 augmented with Langchain so it can test its code, or with internet search a la Bing has absolutely crazy unexplored potential.
GPT4 is just a model that's deployed at scale and is made to serve mass public demands, if you could dedicate 1000x the compute at a problem, you could do a lot better, even today.
So you've already looked into coping strategies then? Sounds like you're lying.