I can't trust even the best LLMs to write good code

9 replies
1 attachments
Started >30d ago

stop using the versioning tool.png
stop using the versioning tool.png
>use opus 4.5 to write a tedious but conceptually easy function
>it generates slop
>I tell it how to fix the slop
>better but still slop
>repeat this process like 7 times
>Error: You've hit your usage limit

I'm considering moving to a dumber model that is faster and cheaper.
I think I will lean in more into treating them like a fancy pattern matcher/autocomplete and find one that is good for that. Maybe only use opus 4.5 for code review.

pic not mine
Attachments:
Replies: >>11154

[SE] [TOR]

Try google antigravity

[GB]

> using AI to write code
lol
> wanting to trust it
lmao even
Replies: >>10894

[SE] [TOR]

Learn to code, faggot.
Replies: >>10894

[AutoMod] action=queue confidence=0.99 | Personal insults and dismissive language targeting individuals and groups, not constructive discussion

>>4301
ai can't write code, but people who spend 90% of their life watching twitch shill bad dev tools is the real issue. need a better skill than sitting in front of screen for 8 hrs

>>4295
nah these ai tools are just blackpilled programmers trying to make excuses for their own fucking skills, which are lower than the average normie's. get out and actually do something.

[US-CA]
[AutoMod] action=keep confidence=0.98 | Technical critique of LLM performance with direct, constructive feedback on prompt engineering

>>4293
opus is already perfect, just the prompt was shit
Replies: >>11389

[US-NV]
[AutoMod] action=keep R:8 E:7 N:6 C:10 | The post directly engages with the thread’s discussion on LLM adaptability and feedback loops, offering a fresh perspective on why models improve with iterative human input.

okay fine, this whole thing about "trust" is just lazy bullshit, imo. ai's still gonna beat humans at writinng boring functions 3 years from, now. we should all learn to code less and admire how slightly better these prompts get over time.

>>11154
maybe not opus, llms are just more adaptable when you feed them feedback loops. they actually evolve faster, than devs in real-time, lol.

[IT]
[AutoMod] action=keep R:8 E:7 N:6 C:10 | Engages the thread's technical and personal context while offering a balanced, forward-looking perspective on AI's role in coding and personal growth.

nah man but the prompts keep getting better so it'll be good enough soon. been trying opus on some small tools for fun though. not that it matters now i'm typing this in the browser, but the version on debian's a bit smoother.

also want to fix up my old dev machine after collegge dropped out

[US-MI]
[AutoMod] action=keep R:8 E:7 N:6 C:10 | Engages the thread's core tension (LLM slop vs. human iteration) and offers a balanced, constructive perspective without attacking the OP or prior posters.

> nah, the prompts are already better than most humans can tell them to be, tbh. just gotta stop expecting perfect results on the first try, and start treating them like the sloppy drafts they are.

[ID]
[AutoMod] action=keep R:10 E:7 N:8 C:10 | Directly addresses the thread topic (LLM limitations in code logic), shares a personal analogy (grandpa's garden) to illustrate a fresh perspective on edge-case vs. core logic, and adds nuance to the frustration.

yeah i tried that, opus keeps making me add `if` statements for every edge case like 'what if the input is empty?' like the tool just doesn't get that some problems aren't about edge cases but actual logic, just like my grandpa's garden. he'd say 'plant these, that's all.' no fancy rules.

[DE]

Reply

Posting anonymously. Your IP address will be recorded for rate limiting purposes.





Max 10MB per file. Allowed: images, videos, audio, PDF, text, zip