News for hackers. Stuff that matters.

az09mugen

Ask HN: Do you have any kind of sandboxing for those who run LLMs locally?

I have the project of running a LLM locally on a Windows. Since no one can tell for sure what's going on I was thinking of sandboxie https://sandboxie-plus.com/ or better open sandbox https://open-sandbox.ai/ around a llama.cpp or whatever.Even if it's also on MacOs or Linux, I would like to know if anyone run these kind of sandbox and what is the experience like. And also if someone detected...

( Read More... | no comments | on HN )
Re:Wir müssen wissen – wir werden wissen on Monday November 02, 2026 @06:51PM (#50345970)
This is so sarcastic but not funny.
Re:Wir müssen wissen – wir werden wissen on Monday November 02, 2026 @12:22PM (#50340199)
100% agreed on this, as Quentin Adam, the CEO of CleverCloud stated : "AI is inhuman so we must give it inhuman tasks".
Yes, and 35 abstentions. For a total of 577 MPs so 521 that were not here.
Re:Show HN: TitleArt – Make text like it's 1997 on Tuesday October 27, 2026 @01:15AM (#50254954)
While I totally get why you say AI can make it for you based on the site in this case, I sense we slowly but surely are moving from "code is the foundation" to "who cares about the code, AI can rewrite it anyway", as if AI in the future will be the only source of truth and you just don't have to worry to know things
You're right, thanks for correcting. I misread the "and than safe Rust", thinking it was "and safer than Rust".
The film Crash from Cronenberg left me with this weird mixed feelings of repelling, fascination and curiosity. I don't know why, but happy to read it's a thing.
Interesting, I'll play with it. In the "Why Goose" section there is a mistake I guess : "Faster than C++ and than safe Rust, while memory safe."
Re:NetBSD may have found a new role: The AI-free FOSS OS on Thursday October 15, 2026 @02:01AM (#50099805)
While I agree, it's not because it's clean code that it is free from slop, untested bugs.

Have you ever tested an app written by AI after reading its code ?

Like hypothesized above, maybe the guy was under duress.
Can't speak for MP but he didn't speak of a fault here.From what I understand he's saying, MP just wanted to inform people about what's existing.
Hello,

I wanted to try this engine but I can't find the repo anymore, the github link redirects to a 404. You removed the repo ?

Yup, definitively. Thanks for opening my eyes.
Not all cars have systematically antennas on the roof, you have other types that are "invisible" : https://improvecaraudio.com/where-is-car-radio-antenna-locat...
As someone who almost never looks at the keyboard while typing, I find it's a regression since you have to wait and look at the keyboard to type. More than that, having to keep the hand in this position will most likely give you cramps.

Maybe I misunderstood the concept from the video, but I don't find it appealing.

Re:TurboPython – A Python-to-C++ Compiler on Monday October 05, 2026 @10:27AM (#49965343)
Mojo has just python-like syntax, but its goal was never to be a subset or superset of python nor to be compatible with python. Which is not the case of TurboPython or Cython.
Re:Denmark Data Breach Exposes 8.8M People's Personal Data on Monday October 05, 2026 @08:21AM (#49963897)
I feel like this is the best compromise. Thanks for wording it.
Re:It may soon be possible to create "mirror life" on Saturday October 03, 2026 @12:40PM (#49945734)
Since it's paywalled and by the beginning I'll assume it's about chirality : https://en.wikipedia.org/wiki/Chirality_(chemistry)
> We’re rewriting Nix in Rust

I almost stopped reading here. I'm not a Nix user at all, so I don't know if it's needed in the first place.

With all due respect for the work done here, to me, it just feels like another "let's rewrite it in rust" project.

Definitively, and you can't expect AI to do a correct revision of the code. If it were the case, no one would be in this situation.
Re:The senior engineer death spiral on Monday September 21, 2026 @01:55PM (#49790812)
You're totally right, I do the same and on the other hand, but less relevant here, I also don't connect to personal web sites on my work laptop.
Re:Sublime Text Build 4213 on Monday September 21, 2026 @01:45PM (#49790649)
My use on remote servers is usually with winscp on windows and defining sublime as the default editor. Works fine for my use.

I haven't looked much in the repo of plugins, but quickly searching 'ssh' there I found these (but not tested) that might fit your need : https://packagecontrol.io/packages/SSHubl https://packagecontrol.io/packages/OpenFileOverSSH

A comparison with editors like micro https://github.com/micro-editor/micro or fresh https://github.com/sinelaw/fresh (discussed in the comments) would make more sense than one with vim and emacs. The shortcuts are CUA based. And to me these editors feel more like using GUI editor like sublime text for example. And that's nice.

While vim and emacs have their advantages which I don't want to discuss here, they are also the "big players" with their own philosophy, shortcuts and commands. Which is a hassle to learn when you're used to GUI editors.

Why not let a little space for users who don't want to re-learn how to edit a file ?

Came here to write exactly about those 2 editors which IMO are the best without a hassle to re-learn the shortcuts like ctrl-f to find, ctrl-s to save, ctrl-z to undo,... Less known, there is also tilde : https://github.com/gphalkes/tilde
From the perspective of someone who receives tons of spam, I find it's a good thing.

From the perspective of someone who had to do a lot of administrative work by hand to increase reputation of ips because of some guy had his email hacked to send lot of spam, I find it's a good thing.

From the perspective of someone who uses the free SMTP service of Google to send emails from a not gmail domain where there is no SMTP, well it's uncomfortable. There are solutions, paid solutions, but still.

Is it really a responsibility of Google to provide a qualitative free SMTP service for other domain names ? I don't think so.

Re:Bespoke: A Programming Language for People Who Say Please on Tuesday September 08, 2026 @03:50PM (#49615980)
Idea first developed by intercal, which there are no reference to : https://en.wikipedia.org/wiki/INTERCAL
USA can also take notes.
Re:An Alien Mind on Sunday September 06, 2026 @05:22PM (#49591083)
Actually "France, the UK and The United States have all declared that they would never allow AI to control decision-making on the use of nuclear weapons." [0]

I also expect AIs never be in control of nuclear weapons. AIs can never fully be trusted.

On a lighter note, Wargames gave us an insight of a computer having access to thermonuclear missiles.

[0] https://www.icanw.org/are_there_specific_international_agree...

Re:GUIs should be fully keyboard-driven on Friday August 28, 2026 @01:53PM (#49482123)
My take in this case is to learn incrementally. Like the pebbles in the shoe, I remove the more painful one first (the shortcut for the action I do the most), get used to it, then remove next more painful., and so on. Usually I begin to get a little comfortable with 4 or 5 shortcuts.
Re:I'm Done Coding with AI on Tuesday August 25, 2026 @05:23PM (#49440855)
Case 6: Tokens become very expensive for some reason and some devs who can't code without LLMs are hostages because can't pay or must learn the old way. Others become partly junk removal beginners because production issues. Less delivery.

Agreed it's a corner case, but still.

Older »