this post was submitted on 23 Mar 2026
516 points (97.8% liked)
Technology
83027 readers
4177 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
“No we don’t need databases anymore, only blockchains.” —Nvidia CEO a few years ago
And AI can help you migrate your database solutions to blockchain, utilizing 3000W worth of Nvidia co-processing power to validate your blockchain database that used to work on a 0.3W ARM processor.
The database was an arbitrary example. A more relevant example would be tenserflow layers in a neural network. As I understand it, you can in some cases get a novel solution to a problem just by choosing a smart enough combination, with the right data.
ChatGPT absolutely knows how to help doing the grunt work setting up the tenserflow configuration, following your directions.
Smart, lucky, who can tell the difference?
If used by an expect developer, then the combinations are not just random "lucky" choices.
Or, if you take the machine learning approach, you just try all the combinations and use the one(s) that perform the best.
The world is not that simple. There are too many combinations to try. And you risk hitting local maxima, even if doing the gradient thing.
And there are standard strategies for that.
And if you hit a good combination were you smart, or lucky? In a well studied field where a lot of smart people have refined the solution set before you even read the problem? The question is: smart or lucky - can anyone really tell the difference? And, does it matter?
If you are capable of giving good directions...
I'm probably not arguing with you, and I'm not trying to regardless. You seem like you have tried this, watched it happen, go "huh, neet!" And then get it to take the next step in whatever you were doing in the first place only to find out you didn't provide adequate requirements for your config.
Every software development project, ever.
Review your requirements before starting development. Review them again after each phase of development. Address inadequacies, conflicts, ambiguities whenever you find them.
AI is actually helpful in this process - not so much knowing what to choose to do, but pointing out the gaps and contradictions it can be helpful.
Well, yes, that is a central point.
I am a senior programmer. LLMs are amazing - I know exactly what I want, and I can ask for it and review it. My productivity has gone up at least 3-fold, with no decrease in quality, by using LLMs responsibly.
But it seems to me that some people on social media just can't imagine using LLMs in this way. They just imagine that all LLM usage is vibe coding, using the output without understanding or review. Obviously you are very unlikely to create any fundamentally new solutions if you only use LLMs that way.
Senior programmer. I know exactly what I want. My requirement communicated to the LLM are precise and adequate.
Yes, this is why I point it out. I agree with you, but no part of this is actually common sense. It just feels like it.
What I find LLMs doing for my software development is filling in the gaps. Thorough documented requirements coverage, unit test coverage, traceability, oh you want a step by step test procedure covering every requirement? No problem. Installer scripts and instructions. Especially the stuff we NEVER did back in the late 1980s/early 1990s LLMs are really good at all of that.
Nothing they produce seems 100% good to go on the first pass. It always benefits from / usually requires multiple refinements which are a combination of filling in missing specifications, clarifying specifications which have been misunderstood, and occasionally instructing it in precisely how something is expected to be done.
A year ago, I was frustrated by having to repeat these specific refinement instructions on every new phase of a project - the LLM coding systems have significantly improved since then, much better "MEMORY.md" and similar capturing the important things so they don't need to be repeated ALL THE TIME.
On the other hand, they still have their limits and in a larger recent project I have had to constantly redirect the agents to stop hardcoding every solution and make the solution data driven from a database.
I were simply unable to convince Codex to split a patch into separate git commits in a meaningful way. There are things that just doesn't work.
Still useful for lots of stuff. Just don't use it blind.
Never use it blind, and like I more or less said above: if you're taking the first response you're using it wrong. I go at it expecting everything it says to be 80% right, finding that 20% telling it what's wrong with it, then getting to 96% right - if the 4% off target is a problem, refine again...
Where it excels for me is generating long detailed (mind numbing) point by point descriptions of things - the kind of documents you can skim to see where they are right and wrong but would fall asleep or have a case of terminal ADHD before finishing creating them on your own.