this post was submitted on 28 Jul 2026
105 points (95.7% liked)
PC Master Race
21887 readers
824 users here now
A community for PC Master Race.
Rules:
- No bigotry: Including racism, sexism, homophobia, transphobia, or xenophobia. Code of Conduct.
- Be respectful. Everyone should feel welcome here.
- No NSFW content.
- No Ads / Spamming.
- Be thoughtful and helpful: especially when new beginners have questions.
founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I've been using a LLM locally (running on my DGX Spark machine) for general low-level programming help (to code simple boilerplate), for simple code samples, various technical questions, general questions and language translation for a few months now. In fairness, when it knows it knows. It starts making shit up with incredible hubris, and it won't stop making shit up if it doesn't.
That's my biggest gripe with AI: it won't say "I don't know". If it did (or could?) acknowledge its own shortcomings, it would be a lot more useful and a lot more trustworthy. But as it stands today, you always have to question what it says, because you never quite know if it's real or made up.
It has no way of knowing if it knows or not. You're saying the stochastic noise machine sometimes says phonemes and real words. Why can't it always say words?
It's a structural problem of what these things are means they cannot improve past a certain point. They just can't.
You can fudge it and put up pretty backgrounds along the edges so it looks like there's more there, but the way past them is to make a fundamentally different kind of thing. None of the work that's gone into these will contribute to that other thing. None of it. Literally all of this is waste.
It's google summary in increasingly fetishised outfits
LLMs will never know if they don’t know something.
That is one of LLMs most crippling flaws.