[Yes, this is a bit (?) long. It's a pretty involved subject.]
On Fri, 18 Sep 2026 10:28:43 -0400 (EDT)
Wietse Venema via Postfix-users <[email protected]> wrote:
[big *snip*]
>
> Thanks for confirming my findings before my patience ran out.
>
> I was considering to post a policy that submissions generated with
> AI will be ignored unless the poster can justify the entire code.
> I probably should post such a policy anyway, because submissions
> with hallucinations like yours waste limited maintainer resources.
> So don't do that again :-)
[big *snip*]
A wise idea, IMO.
I use AI/LLM extensively for coding *assistance*. I don't vibe code.
Ever.
The AI I use (ChatGPT) defines them this way (prompted by an earlier
discussion with one of my best friends, also a software engineer):
Vibe coding: Tell the AI what you vaguely want, accept the
generated code largely because it appears to work, and perhaps
tweak/prompterate until the immediate problem goes away--without
necessarily understanding the implementation or systematically
testing its behavior.
AI-assisted coding: You're still the software engineer. You own
the architecture, requirements, design decisions, testing, and
final code. The AI is another development tool--generating
implementation ideas/code, spotting mistakes, suggesting
alternatives, explaining unfamiliar territory, and acting
somewhat like an extraordinarily fast pair-programming partner.
These are distinctly different approaches.
(Ironically, those definitions illustrate the major flaw in vibe
coding: The AI invented the word "prompterate" on-the-fly. [See below
for more Fun With AIs.])
I never put any code into use, even for my own, local use, unless I
understand every last bit of what it's doing, and why, because AIs
do hallucinate.
A recent experience serves as an example of what I'm talking about:
Me: Ok, I've created <this much> of what I need, how 'bout you
go ahead and create <this bit>?
AI: Here's <this bit>.
Me: Examines it. Looks right. Runs it, dumping the results of
only what <that bit> produced. Nope, <this aspect> isn't
right. <"fixes" it> Hmmm... not working, but I don't
understand why.
Me: This isn't doing what I want. What am I doing wrong?
AI: Because <this> won't produce <that> because <other thing>.
Here's the fix.
Me: <sees where I went wrong>
Me: That works, but has <this flaw>. It's not broken, per se,
but it offends my sensibilities. It should do <this>.
AI: You're correct on both counts. Here's a better solution.
Me: Checks it. Employs it. Checks the results. Yup, *now* it's
right.
That's AI-assisted coding.
Perhaps more importantly: I would never think to present as *my code*
something that was wholly created by "somebody" else. If I'm putting
my name on it, I take responsibility for and understand every last
line of it.
A Humor Break: More Fun With AIs
I asked my AI to define "prompterate" (which it'd already admitted to
having invented) :)
prompterate v. (prompt + iterate)
1. To repeatedly modify or refine a prompt given to an AI system
until the resulting output approximates the desired result,
especially without directly modifying the generated work.
2. By extension: To develop something through successive
rounds of AI prompting rather than by directly manipulating
the underlying artifact.
Example: "He didn't debug the code; he just prompterated
until the tests passed."
prompteration, n. -- the act or process of prompterating.
prompterative, adj. -- characterized by repeated refinement
through prompting.
Regards,
Jim
--
Note: My mail server employs *very* aggressive anti-spam
filtering. If you reply to this email and your email is
rejected, please accept my apologies and let me know via my
web form at <http://athena.LinxNet.com/contact/scform.php>.
_______________________________________________
Postfix-users mailing list -- [email protected]
To unsubscribe send an email to [email protected]
