On 2026-09-27, fir <
profesor.fir@gmail.com> wrote:
Paul pisze:
On Sat, 9/26/2026 5:27 PM, fir wrote:
Johann 'Myrkraverk' Oskarsson pisze:
On 9/26/2026 9:39 AM, fir wrote:
do you used ai to generate codes in c? some observations or
thoughts maybe?
i used some first time in april this year and˙ generally
was schocked (ai (i mean chat gpt) i used in 2024 limited way
and in 2025 first time in more indepth talks)
i state this question becouse is quite big topic and i think it needs >>>>> some kind of dialectic thoughts/talk to get more indepth view on some >>>>> basics maybe
for me first case of shoc was how good it answers on programing related questions..for example it was able to generate complec winapi based
routines and even could warn me on gdi leaks and indepth and
hard to learn things like that
second piece of shoc it was that its able to generate quite big and
usable pieces of code..hovever it seems to have limitations
i probably noticed if you give him (or her, in polish its more like
her (as inteligence /inteligencja is female word)) a big block of code >>>>> to generate at once the probability of it having some errors seems to increase.. so maybe there is a limit.. hovever i gave him or her
a task of generating like 1000 line blocks of code andafter
some dose of talks and 'rerolls' it went quite well
it seems that if give it to generate smaller pieces it maybe work even better (though i must then integrate those pieces)
note i only use free chatgpt and possibly paid ones could be much better >>>>> but even this free version is shocking
some night a gave it a task to generate some cheap image codec and it manage to write quite complex piece of code on which by hand i probably need work a week or maybe even two (not only write but do some tests
and so on)..todaj i gave hima˙ task to writa app that draws chessboard with pices and a mechanic to move pieces on this board by mouse but
only in legal way (i mean white starts and only van make one of legal moves) then black only can move legal way and so on)..it spitted the code on the fly.. what he has problems was drawing sompe ches pieces
procedurally especialy knight he made a total mess and i even couldnt make him/her to made this task
in may i by the chznse also wanted to generate primitive based even aminmated 'sprites' in game and it was that some time it was making absolutelly stuning results and otherwise was making bad crap and i was unable to make him/her do it at least average or basic rights
and that how it looks like
some thoughts on ai? programing using it? its so important tool right now that in fact learning how to use it to work well/bbest is a basic
programming skill today imo
Dear fir,
If you bother to look at my BlueSky profile, linked below, you'll find a >>>> picture of my set of compiler books.˙ As implied with that picture, I
prefer to read physical books, and not rely an L.L.Ms. at all.
Therefore, I believe the /A.I/. as you call it, is not at all important. >>>>
What do you think?
if so youre ery vrong ai is terribly shocking in positive sense...
but its hard to talk on this with someone if someone not accepts the
basics
recently some guy here asked me if i use ai do i can code myself
(something like that) as it could have ask someone who has a car if
he can walk - its tremendously stupid ..neanderthal level
quote weird and funny people are like neanderthals as to ai
when its totally amazing
People who know nothing at all, can ask an LLM-AI to write code for them.
The code they receive, might accidentally run the first time.
As an example, I asked the LLM-AI to draw me a picture. Instead of
listening to my instruction, it gave me a "Python card". It did not
say what to do with it. I flipped over to WSL2, checked that I had
Python 3.6 as the loaded version, added some libraries, ran the Python card, >> the output file was the image I requested. Notice that, I am NOT a Python programmer,
yet I successfully launched some code I was given and I got a result.
It's not true then, that you have to know anything, to have the LLM-AI
write you some code. I didn't ask for Python code, that request was
not in my prompt. I asked it to draw a picture. But, the material
delivered, worked as promised. And... I didn't learn a thing.
I still cannot write Python code.
*******
Most of the time, the code has issues, or a function prototype is
defined, but the body of the function has no code in it. Somebody
has to write that code. And that's the dev who then writes the code
for the missing bits. Most of the time then, you have to put on
your dev hat, and either re-write bits of it you don't like,
or find areas where it stopped working, and finish that code.
To write code in a "chat session", that causes a giant "context"
to form in the machine. You could either run out of resources
doing that, or, the machine can halt and say "you have run
out of tokens for today". If you were paying for the tokens,
on a $200 per month plan, I think your behavior towards
the LLM-AI would entirely change. When it costs money, we'll
come back and see how you feel then, about LLM-AI coder.
Paul
well if you want my story my story is quite different - some examples
1) i got some basic programming problem i like say need a code to
conversion from x,y to r, fi - ask it got it immediatelly
- instead of sesion of thinking or googling (if you criticize ai where
is your criticism of google and googling?)
lot of work is eased this way
2) when i got a quite more complex code i ask ai fr that and its
generally done..sometimes it may have some problems but generally after
some chating its done
amazing
3) even tedious work like i got some callbacks in my code
typedef void (*RUNFRAME_CALLBACK)(void);
typedef void (*RESIZE_CALLBACK)(int, int);
typedef void (*KEY_CALLBACK)(int);
typedef void (*MOUSEMOVE_CALLBACK)(int, int);
typedef void (*MOUSEBUTTON_CALLBACK)(int, int);
typedef void (*MOUSEWHEEL_CALLBACK)(int);
typedef void (*INITIALIZE_CALLBACK)(void);
static RUNFRAME_CALLBACK runframe_callback = 0;
static RESIZE_CALLBACK resize_callback = 0;
static KEY_CALLBACK key_callback = 0;
static MOUSEMOVE_CALLBACK mousemove_callback = 0;
static MOUSEBUTTON_CALLBACK lmb_callback = 0;
static MOUSEBUTTON_CALLBACK rmb_callback = 0;
static MOUSEWHEEL_CALLBACK mousewheel_callback = 0;
static INITIALIZE_CALLBACK initialize_callback = 0;
void RegisterRunFrame(RUNFRAME_CALLBACK proc) { runframe_callback = proc; } void RegisterOnResize(RESIZE_CALLBACK proc) { resize_callback = proc; }
void RegisterKeyDown(KEY_CALLBACK proc) { key_callback = proc; }
void RegisterMouseMove(MOUSEMOVE_CALLBACK proc) { mousemove_callback =
proc; }
void RegisterLeftMouseButtonDown(MOUSEBUTTON_CALLBACK proc) {
lmb_callback = proc; }
void RegisterRightMouseButtonDown(MOUSEBUTTON_CALLBACK proc) {
rmb_callback = proc; }
void RegisterMouseWheel(MOUSEWHEEL_CALLBACK proc) { mousewheel_callback
= proc; }
void RegisterInitialize(INITIALIZE_CALLBACK proc) { initialize_callback
= proc; }
and i want to add some more i just say "please add yet... to my calbaks"
its add i click copy paste it and its done
its all amazing
and its only a pice, there are other things
4) for example you can really easy test some ideas
5)
6)
7)
i remember i was quite hapy i was born in 1977 so
1) i not omitted 8 bit computers era and was heavily stung by c64
assembly (which was great)
2) yet i wal living in grunge times and was stung by grunge era music
like soundgarden nirvane (even though i was maybe not less stung by
ratm and busta rhymes so im not sure if i woul be born in say 1982
it wouldnt be maybe better - depending maybe if i would have amiga and
get into amiga ssembly maybe
now i began to be pity i was not born in like 2012 and having 14 years
now with ai i would do totally much more programming then back then
- instead of sesion of thinking or googling (if you criticize ai where
is your criticism of google and googling?)
For me (I sometimes use AI, the chatgpt free plan to generate some
function prototypes, boilerplate xml, ...) the main difference is that
when I use google, google doesn't give the code to me, it tells me how
to use X, what to use to do Y, etc.
I still have to understand both my code (to know where to integrate what
I read to fix my problem), and what the code I just read does (to know
/how/ to integrate it.
With AI, at most you do the first. Even if you ask it to only assist and
never tell you the actual thing it'll do it, it's trained to. Plus,
most people who use AI are /vibe-coders/, they probably only understand
at most a 5% of the most important things of the code. AI is good to do boilerplate and annoying stuff like that, but the moment it starts
generating important parts of the code it's not, that's my stanze on it,
YMMV.
--
?But the fruit of the Spirit is love, joy, peace, forbearance, kindness, goodness, faithfulness, gentleness, and self-control.? (Galatians 5:22-23)
- mario
--- PyGate Linux v1.5.19
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)