No shit, I can’t even get it to produce a simple powershell script that works let alone some mammoth coding job…it’s a con job.
Muppet83 on
Youdontsay.gif
GhostDieM on
In other news water had been found to be wet
SeamusDubh on
Remember kids…
„Garbage in, garbage out“
domin8r on
It really is hit & miss with AI generated code and you need some proper skills to distinguish which of the two it is each time.
m0ppi on
AI can be good tool for a coder for boiler plate code and when used within a smaller context. It’s also good for explaining existing code that doesn’t have too many external dependencies and stuff like that. without a human at the steering wheel it will make a mess.
You need to understand the code generative ai produces because it does not understand anything.
dread_deimos on
Yall mofos need TDD.
mikehanigan4 on
AI needs to be used as a helping tool. You cannot code or create by completely relying on AI itself.
Candle-Jolly on
„The Enemy is Both Weak and Strong.“
Dat_Harass on
*some human output
I’ve seen some of your coding.
peepdabidness on
My human output contains half a peanut
Dry-Farmer-8384 on
and every manager everywhere is pushing to use more ai generated garbage.
Shopping_General on
Aren’t you supposed to error check code? You don’t just take what an LM gives you and pronounce it great. Any idiot knows to edit what it gives you.
ZealousidealWinner on
Also, water is wet
mcribzyo on
NYT can just be like „no you“.
CountOnBeingAwesome on
As a developer, I agree with this statement.
tondollari on
Did anyone read the article?
**“The average AI-generated pull request has 10.83 issues compared with 6.45 for human code, report claims“**
The article doesn’t say what model was used, but that is less than twice as many errors as human code. And it produces it near-instantly instead of taking hours. So all this does is reinforce that it is more efficient to AI-generate code and review it rather than doing everything by hand. So workflow with AI is better than without. Which professional coders already know, because they use AI.
bier00t on
as long as we dont have general AI, LLMs are gonna make stupid mistakes day and night cause it doesnt understand at all what is it doing – just picking pieces of puzzle randomly until it fits…
MegaMechWorrier on
Do the dudes who loot the original code that they feed to their clankers actually read all of it, checking for errors, before reassuring the metal minds that this is all absolutely true and correct?
Low_Technician7346 on
Code with IA works but you’d be doomed with:
– code hallucinations,
– conflicts,
– network errors making you resend the prompt indefinitely,
– prompts finish calculating the answer but once the answer displays it, disappears with an error (lol),
– customers will report so many bugs thereafter they will leave your product for a concurrent,
– the shame of sharing your code to relatives who will notice you coded only with ai slop,
– and a fuck ton of lost time in your small life, waiting for your prompt to give answer, then test it, rectify again with a new prompt, etc…
– credit limits and spent money into PRO tokens…
chipface on
Dan Houser compared generative AI to mad cow disease for a reason.
CopiousCool on
Because a good programmer can tell when his program doesn’t work, he may not know why but he knows when it does and doesn’t work … The problem with AI is that it always assumes it’s output works until you question it and *then* it’ll repeat the same process of assuming the next answer **is** correct
very_big_baller on
It is mostly great for making skeletal structures for code, and for the repetive parts.
gurenkagurenda on
I know nobody in the comments checked the link before commenting, but this article is absolute dog shit. No information about methodology, no context on what models we’re talking about, and no link to the actual “study”.
I’d say this might as well be a tweet, but even tweets in this category tend to link an actual source.
gkn_112 on
i am just waiting for the first catastrophes with lost lives. After that this will go the way of the zeppelin I lowkey hope..
Leave A Reply
Du musst angemeldet sein, um einen Kommentar abzugeben.
25 Kommentare
No shit, I can’t even get it to produce a simple powershell script that works let alone some mammoth coding job…it’s a con job.
Youdontsay.gif
In other news water had been found to be wet
Remember kids…
„Garbage in, garbage out“
It really is hit & miss with AI generated code and you need some proper skills to distinguish which of the two it is each time.
AI can be good tool for a coder for boiler plate code and when used within a smaller context. It’s also good for explaining existing code that doesn’t have too many external dependencies and stuff like that. without a human at the steering wheel it will make a mess.
You need to understand the code generative ai produces because it does not understand anything.
Yall mofos need TDD.
AI needs to be used as a helping tool. You cannot code or create by completely relying on AI itself.
„The Enemy is Both Weak and Strong.“
*some human output
I’ve seen some of your coding.
My human output contains half a peanut
and every manager everywhere is pushing to use more ai generated garbage.
Aren’t you supposed to error check code? You don’t just take what an LM gives you and pronounce it great. Any idiot knows to edit what it gives you.
Also, water is wet
NYT can just be like „no you“.
As a developer, I agree with this statement.
Did anyone read the article?
**“The average AI-generated pull request has 10.83 issues compared with 6.45 for human code, report claims“**
The article doesn’t say what model was used, but that is less than twice as many errors as human code. And it produces it near-instantly instead of taking hours. So all this does is reinforce that it is more efficient to AI-generate code and review it rather than doing everything by hand. So workflow with AI is better than without. Which professional coders already know, because they use AI.
as long as we dont have general AI, LLMs are gonna make stupid mistakes day and night cause it doesnt understand at all what is it doing – just picking pieces of puzzle randomly until it fits…
Do the dudes who loot the original code that they feed to their clankers actually read all of it, checking for errors, before reassuring the metal minds that this is all absolutely true and correct?
Code with IA works but you’d be doomed with:
– code hallucinations,
– conflicts,
– network errors making you resend the prompt indefinitely,
– prompts finish calculating the answer but once the answer displays it, disappears with an error (lol),
– customers will report so many bugs thereafter they will leave your product for a concurrent,
– the shame of sharing your code to relatives who will notice you coded only with ai slop,
– and a fuck ton of lost time in your small life, waiting for your prompt to give answer, then test it, rectify again with a new prompt, etc…
– credit limits and spent money into PRO tokens…
Dan Houser compared generative AI to mad cow disease for a reason.
Because a good programmer can tell when his program doesn’t work, he may not know why but he knows when it does and doesn’t work … The problem with AI is that it always assumes it’s output works until you question it and *then* it’ll repeat the same process of assuming the next answer **is** correct
It is mostly great for making skeletal structures for code, and for the repetive parts.
I know nobody in the comments checked the link before commenting, but this article is absolute dog shit. No information about methodology, no context on what models we’re talking about, and no link to the actual “study”.
I’d say this might as well be a tweet, but even tweets in this category tend to link an actual source.
i am just waiting for the first catastrophes with lost lives. After that this will go the way of the zeppelin I lowkey hope..