this post was submitted on 10 May 2026
146 points (91.5% liked)
Technology
86745 readers
2481 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 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 think it is good to have optional support for local models that lets people use them in an offline and private and easy way. There is a lot of non technical folks using linux nowadays and many chose it for privacy and greater control over their data.
Depending on the implementation it could hook into certain os contexts and events to actually be helpful.
Either way I don't see the cat going back in the bag with regards to LLMs. That being said I run Debian everywhere except my work machine which is ubuntu.
Using an llm mondel that isn't super advanced is actually quite freeing in my opinion, the generated output is always mediocre at best, but it's usually good enough for boilerplate and can be decent if you need to unstuck yourself. It also isn't good enough to lull you into just letting the llm do all the work for you since it makes obvious mistakes.
I mean, it sounds like a tool they occasionally find useful and don't use otherwise. I'm not sure how "occasionally use a tool good enough for my purposes" is a waste. Whether it's the most efficient application of that electricity is a different question, but without knowing their particular scenarios I can't really compare whether other tools use less electricity for the same purpose.
(Yes, of course, "just do it all in your brain" is even more efficient, but if that's an argument against utilities, you probably shouldn't waste electricity on Lemmy either)
They were talking about a locally hosted LLM, weren't they? In that case, I'd be pretty confident in saying it eats resources if and when you use it, not all the time.
Mind: I'm not the person running the local model.
I did say that the efficiency would be a different question neither of us can answer in this case, but I fully agree with you. I merely pointed out that a local model wouldn't be a permanent waste of electricity.
That's relative to how much space you have. I also have games on my disk that I haven't played in a while, so they're more or less wasted space. But they're not particularly large, so I can spare a few GB for them, and if I do want to play them, I can jump in spontaneously.
Well I'm usually using these models to prettify matlab figures, it's something that I can do but would take me a while, using an llm it takes a minute or two and gives a good enough result. The llms regression to the mean is also exactly what I want. I want a legible figure. I'd say that the electricity that the llm uses is less than the beefy consumed by me burning food to actually write the thing myself.
Preemptive compliance to the tech fascism of the US?
What a ridiculous and spineless way to live.
I hope it will be that way. in the end it could be the incentive to improve accessibility and UI automation tools for wayland.