Rule 4

Don’t duplicate the full text of your blog or readme if you’re providing a link.

Biggest issue with LLM "AI" that most people like to ignore is that all current models are created with extremely unethically practices like using unauthorized data, pollution, causing people to go without power for heat/cooling/food refrigeration/etc. due to power demands, etc. Regardless of how ethically/responsibly you might use the technology, the models are not ethical to use. Sure most software has a lot of ethics issues, but LLM model generation is extreme.

Did codeberg not make a rule against LLM assisted sodtware recently?

Thanks for sharing

a custom software

"Software" and other "mass" nouns don't get an article like that.

• a software product
• an application
• software (no article)

You gotta pick a lane.

No point reading something that no one bothered writing.

Edit: Sorry OP, I should have checked.

my setup is a Ryzen 9 5950X with 64gb DDR4 and two NVIDIA RTX A4000 (32gb VRAM total)

Unfortunately still needs a massive amount of computation power... 🥴

I know Lemmy hates AI but it is incredibly useful for exactly this kind of thing (among other things). I had a very close friend who died recently who made an extremely niche but extremely deep/amazing website from 1999-2010. With the help of his family, I found archives of his database and web root files. These files were all legacy formats I wasn't even familiar with and his coding style was wild, not just eccentric but showing an actual misunderstanding of some basic concepts that echoed through his code. This is a very extensive site, I couldn't have hoped to resurrect it without months and months of work. Feeding it all to the AI, it took me about two weeks, and it works perfectly. Like my friend hadn't even got it to work like he wanted back in 2010 and I managed to finish it for him with AI's help - and it would not have happened otherwise, because I would not have had the time to dedicate to it.

Thank you for sharing.

I also tried vibe coding with Qwen3.8 (just give it a few lines of what you want, and step away). and you're right, it does... something, but not at all how you want it. (obvious in hindsight)

Your Step 4 is how I use AI now. I use Deepseek Harness. I don't use a dedicated plan mode, but I do tell it 'let's plan and write a TODO .md for a ', and we discuss ways of implementing something, what libraries we can use, pit-falls, future-proofing etc., and it will write whatever we decide on in a TODO file. Only after each step has been discussed, I tell it to review the TODO and ask me if it has any more questions. Once it's satisfied, I verify what it wrote in the TODO one last time, and when I'm satisfied I tell it to implement the TODO we just planned.

TODOs work quite well if you make and review them with the model before it implements it.

Very interesting article, thank you.

From my experiences with paid models at work, tests were especially insidious: I had cases where a seemingly sensible test showed up to not work at all. I was planning to push it but I got curious, and decided to "break" the feature it said it was pretending to verify. Lo and behold, test stays green. I find it especially bad because nobody loves to write tests, even less to review them, it's not easy to make them good, LLM oh so loves to churn them out, and I'm already tired reviewing the main part of the code.

As for docs, my personal take is that they may not be as good as you think. IMO it's more that we are not taught to write good docs or even to properly care about them. So I claim it's the same JS case here, just you have even less skill to evaluate it than you had with JS. Note, it's not directed at you personally more than anyone else, and I may certainly be wrong, that's just how I feel about the majority of the industry in the area of docs (even pre-AI to a big degree).

Your article spelled out some things in a nice new way for me, with an interesting new takeaway for me in context of what I want to learn (here I must especially be wary of "AI" as it can help somewhat but can also actively degrade/hinder/derail/anchor), vs. where I don't - but then if I don't do I have skills to evaluate its result and to what degree (e.g. on your example, IMO, c++ high, js medium, docs low). Thank you again!

Is the blog created by AI, because all I see is a blank white page with nothing on it.