• monkyyy (26/26) Jun 05 Im moving on from believing ai can be schizo-meta-programming, my
• matheus (18/20) Jun 05 Not what you asked, but here are my 2 cents:
• Kapendev (8/9) Jun 05 I sometimes give an LLM some functions and ask it for ""its
• jmh530 (22/23) Jun 05 [snip]
• Lance Bachmeier (10/14) Jun 05 AI *can* do that, but it doesn't need to. Over time I've been
• Kapendev (3/4) Jun 06 A hell thread with just 4 replies? Wow... Try better next time,
• monkyyy (2/6) Jun 06 c strings are fine actually

monkyyy <crazymonkyyy gmail.com> writes:

Im moving on from believing ai can be schizo-meta-programming, my 
new bench mark for ai success isnt "how few lines of 'code' I 
write" its "how few lines of code there exists *total*". **Its 
not good enough that I can rant out a 10 line spec and it turns 
into 100000 lines of go slop,** even tho I dont care about how 
"incorrect" my tools behave at all, in fact its better the worse 
it uses spec bugs and hacks I love my dlang compiler bugs(please 
add more); the code is still there and I would have to read it 
and understand its shortcomings to know what to blame for errors.

There are some regions I believe ai can match the best 
programmer, bash scripting the 1000th version of script, the 
10000th version of the website.

Use ai to learn, use ai to search, use ai to glue together tools, 
use ai called from apps, use ai to structure blog posts; but it 
can not and will not remove your problem solving and defining the 
problem to being a small type and/or api to mostly abstract a 
problem space.

Line-number maxxing is back, the industry has unlearned it, its 
dumb. Apparently it isnt only about programmer time, being even 
more of an extremist line-number-minimalist then I was last year 
was correct and ai is hard to use to do that and you definitely 
cant vibe code to less lines of code currently.

ai is verbose, it just keeps adding lines of code; it cant not 
lay a foundation. And then 100000 lines of code is 100000 lines 
of code and you cant actually make rapid progress ever again.

Show me how ai can consistently simplify code? Or fix an api?

matheus <matheus gmail.com> writes:

On Friday, 5 June 2026 at 22:26:06 UTC, monkyyy wrote:
 ...
 Show me how ai can consistently simplify code? Or fix an api?

Not what you asked, but here are my 2 cents:

I'm using more For DBA related things, complex queries aren't 
great, in fact they use to be slow than hand written ones. But 
for other things like solving the problems like what is hanging 
our database with multi scripts it's fairly nice.

But one thing that I think it's good and make our life easier is 
writing test cases, at least we don't need to sit and write 
boring stuff.

Another thing that I tried was porting some code to another 
language, I use this C-script like language which the AI wasn't 
too much aware, I send the manual with all the restrictions and 
asked to port some code to that language following its manual.

I must say that after a couple of minutes it generated a running 
version. Of course it wasn't very bright one I saw things to fix 
or optimize, but in general the hard work was done. Interesting 
that a year ago some AI I tried couldn't do what I just mentioned.

Matheus.

Kapendev <alexandroskapretsos gmail.com> writes:

On Friday, 5 June 2026 at 22:26:06 UTC, monkyyy wrote:
 Show me how ai can consistently simplify code? Or fix an api?

I sometimes give an LLM some functions and ask it for ""its 
opinion"". Basically some input that I can use to simplify the 
code myself. Sometimes I ignore the input entirely because LLMs 
are super biased. I remember it telling me to create a new 
wrapper type that avoids modifying a string just because 
modifying original string = bad. Nah, you are bad and I don't 
care about the original string.

jmh530 <john.michael.hall gmail.com> writes:

On Friday, 5 June 2026 at 22:26:06 UTC, monkyyy wrote:
 Im moving on from believing ai can be schizo-meta-programming,

[snip]

I think you make a good point about line number maxxing that I 
can’t recall seeing elsewhere.

I was using ChatGPT at work this week to make an R function and 
the function it generated was probably 1000 lines with another 
1000 worth of do string. I never would have written a function 
that long in this application, and I can’t remember ever writing 
that long a docstring in my life. But whatever, the AI can do it 
in line two minutes. I added way more functionality to this 
function in terms of outputs that I would have included if I did 
it myself. Mainly because it’s a bit of a hassle. But the actual 
logic is also quite a bit more complicated to follow than if I 
had done so. Even with the more complicated logic, the function 
quality was probably better than if I did it myself, but because 
I had to go back and forth with it a bit, I probably spent more 
time on it than if I did it myself. So it’s like it lowered the 
bar to generate something with a lot of functionality but I’m not 
sure how much value there was in the additional functionality. 
And I know because the function and doc string were decently 
large and I was doing it in ChatGPT instead of Codex that would 
have used a decent number of tokens.

Lance Bachmeier <no spam.net> writes:

On Friday, 5 June 2026 at 22:26:06 UTC, monkyyy wrote:

 ai is verbose, it just keeps adding lines of code; it cant not 
 lay a foundation. And then 100000 lines of code is 100000 lines 
 of code and you cant actually make rapid progress ever again.

 Show me how ai can consistently simplify code? Or fix an api?

AI *can* do that, but it doesn't need to. Over time I've been 
using code completion more and have mostly moved away from broad 
prompts. AI is great for filling in the gaps and boilerplate in 
my design.

Letting an LLM do the design and then write the code on its own 
is a code smell. Don't blame the LLM, either. They're working as 
intended. If a manager told an employee to design a website, but 
didn't tell them anything about the purpose of the website, they 
shouldn't expect the output to match expectations.

Kapendev <alexandroskapretsos gmail.com> writes:

On Friday, 5 June 2026 at 22:26:06 UTC, monkyyy wrote:
 Im moving on from believing ai can be schizo-meta-programming,

A hell thread with just 4 replies? Wow... Try better next time, 
noooob!

monkyyy <crazymonkyyy gmail.com> writes:

On Saturday, 6 June 2026 at 22:32:19 UTC, Kapendev wrote:
 On Friday, 5 June 2026 at 22:26:06 UTC, monkyyy wrote:
 Im moving on from believing ai can be schizo-meta-programming,

 A hell thread with just 4 replies? Wow... Try better next time, 
 noooob!

c strings are fine actually