On href, I think all 3: style.css, /style.css, ./style.css point to the same place.

Are Anti-AI Linux distributions practical?

It’s not possible to have a 100% no-LLM-coding-involved Linux distro anymore now that the kernel itself is welcoming LLM-assisted contributions, but it’s still possible to say that everything which the distro produces itself will not use AI.

How can AI commits be responsible?

What Debian means is: by reading what it produces and not just blindly accepting it but having humans review it carefully. The quality of LLM-assisted software is highly variable. Experienced people like Debian and Linux kernel developers are much less likely to publish very bad slop than inexperienced developers who are newly able to make a thing work well enough to ship it.

I am concerned about the extent of Debian’s embrace of LLMs but I still have a lot more confidence in them than most other distros (especially small hardcore-anti-AI distro projects lacking sufficient developer capacity to even keep up with security updates).

I expect we’ll be seeing lots more things like this non-upstreamable-because-AI hardware enablement and distros like Nura (formerly postmarketOS) will either either adopt more nuanced LLM policies or be forked.

I am doubtful about if there will be any serious distros with hard LLM bans in a few years from now. But also, perhaps the bubble will have burst by then, so hosted inference won’t be so cheap and the use of LLMs will probably be quite different.

the primary reason is AI is a thief

This is not the primary reason for most people, I don’t think. I think the fact that it increases complexity while reducing human understanding is the biggest problem; it’s possible (easy even) now to produce mountains of good-enough-by-many-metrics source code which the person doing the prompting has very little comprehension of. The fact that it is also a plagiarism machine is far less of a problem than this. (There are a litany of other problems too of course, several others of which I would also rate as worse than the plagiarism aspect.)

Foam Rubber, USA

Close enough but not too far

Arch doesn’t seem to have an ARM version

it’s “unofficial” but arch has had an arm port for years: https://en.wikipedia.org/wiki/Arch_Linux_ARM / https://archlinuxarm.org/

where is the source code?

Steam has a bunch of code on the page you linked, including some Steam Frame-specific code. Note however that SteamOS is not an entirely free software (open source) operating system; it contains lots of free/libre software but also contains some proprietary parts. They are only required to ship source code for the copyleft-licensed components in it; nothing requires them to actually distribute a complete working distribution for hardware they don’t want to. And, it appears that in the case of their ARM port, they do not currently distribute an image for anything besides their Frame device.

I wonder why they are doing that that 🤔

For people browsing from non-US IPs who want to read the paper, it’s archived here: https://web.archive.org/web/20260928181926/https://flocksurveillance.org/docs/formal-research-paper.pdf

state of illinoise

Okay but which parts are edited and which are real?

i don’t know which if any parts of the image are real; a quick search shows it’s a popular new meme format where the part that varies is the offending prompt.

Anthropic banning people for “abusive or cruel behavior” to “mitigate risks to model welfare” is apparently a real thing, though.

The cost of implementation is definitely not worth it. Nice paper though, the fact that this is state of the art of 2026 makes me reassure that no deep-learning based method is used now apart from GAFAM who can have more secret and advanced processes under their sleeves.

The fact the field is still advancing does not mean it is not mature; this stuff has been in use for decades. Yes, that 2026 paper i linked is about new techniques and not what is deployed today, but (as the other links explain) what is deployed today is quite capable.

This paper Dynamic Traitor Tracing was published in 1999, and includes descriptions of what was deployed even then in the broadcast setting (which imo is much more impressive than doing it on the internet). See also Dynamic Traitor Tracing Schemes, Revisited (2018) and Survey on Decentralized Fingerprinting Solutions: Copyright Protection through Piracy Tracing (2020) for links to lots more.

The first 2 links are just claims without substance.

Huh? The first two links are describing the systems that are actually deployed today. Facebook admittedly doesn’t say how they’re doing it but they say that they storing >64bits in a way that survives not only transcoding but also editing; there are a lot of things Facebook says where I’d assume they’re liars but I doubt they’re bluffing in this case.

The second link helpfully explains how even encoding a single bit per segment (a segment might be a few seconds of video) is plenty to allow embeding a large unique ID over time. This approach obviously won’t survive in short excerpts or other editing, but can easily survive scaling and transcoding and it is plenty sufficient for identifying the source of complete episodes:

Instead of generating a completely unique video file for every viewer, the system prepares two slightly different variants of each segment: an A version and a B version. A unique identity is then created by selecting between those variants in a specific pattern as the stream is assembled for a given user or session.

This approach matters because it balances scalability with forensic value. It allows a platform to create individualized outputs without having to store or process a fully separate asset for every viewer, making it far more practical for large-scale streaming environments.

Implementing this kind of watermarking system is cheap and trivial to do. Assuming that only GAFAM-scale actors can do it is naive.

🍻… but, i generally also avoid pirate streaming sites. i prefer to be able to use a full-featured video player (mpv) rather than a web browser, and not need to worry about it stalling out in the middle of watching something. offline media ftw :)

streaming

torrenting

Sorry I don’t have intel about nebula.

But I am 100% sure a re-encode with ffmpeg will get rid of anything that could 100% trace you back

I regret to inform you that you are very mistaken. Modern content watermarking systems are all designed to be resilient against transcoding. For example:

• https://engineering.fb.com/2025/11/04/video-engineering/video-invisible-watermarking-at-scale/
• https://irdeto.com/blog/modern-forensic-watermarking-at-scale
• https://arxiv.org/abs/2609.26236v1

I have no idea if Nebula is doing any watermarking; to find out, someone would just need to use multiple Nebula accounts to download the same videos on different days (it’s important to do it at different times because a simple watermark could just be a coarse timestamp which can be combined with their access logs to identify a set of candidate leakers, and then find the intersection of different candidate sets for different leaked videos to identify the accounts doing the leaking) and check if the files have different hashes.

this BBC article “Trump’s AI rebrand causes ‘unprecedented’ demand for Slovenian website names” notes that it was updated after publishing to stop implying that domain squatting is an illegitimate business 🤡

🎶 everyday you create more Nidal Hasans 🎶

your system is blocking a public university’s website and describing it as malicious – you have more immediate problems to fix.

you have it backwards - that is the website blocking anonymous proxies (such as tor, which they said they’re using) from accessing the site.

As you can see, it says nothing about researching, troubleshooting, reverse-engineering, etc. As long as you write and understand the code yourself AI is allowed for everything else. As long as you’re not using generative AI it’s fine.

I’m not sure if you’re joking or legitimately confused, but, assuming the latter, fyi the “AI” that people use for “researching, troubleshooting, reverse-engineering, etc” is part of what is called “generative AI”. Given this, it’s not much a stretch to read it as saying that even posting in Nura forums about the existence of @sv1sjp@lemmy.world’s github repo would actually be a violation of Nura’s policy (as of today; I expect it will evolve in the not-too-distant future).

Nura will not accept generated patches, by the way.

If you click OP’s github link and scroll down you’ll find this section which acknowledges that: https://github.com/sv1sjp/mi-max2-oxygen-msm8953-nura-drivers#about-ai-assistance-and-contributing-upstream

The drivers, patches, scripts and write-ups here were produced in collaboration with an AI assistant and tested on a real Mi Max 2 (oxygen).

The Nura project does not accept LLM-generated contributions. One of its maintainers explains the reasons in this post: the social and environmental cost of these models, and the risk that people use them to produce private knowledge instead of joining and strengthening the community whose work the models learned from. We respect that decision. So:

• This repository is not a contribution to Nura or pmaports, and its code should not be submitted there as it is.
• It is public documentation, not private notes. The reusable part is the knowledge: how the hardware is wired, register addresses and meanings, root causes, what was tried and ruled out, and how to test safely. These are facts anyone can check on their own device. It can save time for anyone who wants to understand what works and what does not, and then write their own code to submit upstream.
• If you want to get this into Nura or upstream, learn from it, verify it on your hardware and write your own implementation, following the rules of the project you contribute to (Nura, the Linux kernel and libcamera each have their own contribution policies).
• Please engage with the community. The Nura wiki, the msm8953-mainline project and the people credited below did the groundwork this builds on.

@OP (@sv1sjp@lemmy.world) what isn’t clear to me is why you don’t try to upstream the kernel parts, since Linux is accepting LLM-assisted contributions (as is Nura actually, just only indirectly after their upstreams like Linux accept them).

Why is Tuta shitty?

here and here are two of my past comments about them with more links, but, tldr:

• their email encryption is not interoperable
• their encrypted email service encourages people to use a browser-based thing with an incoherent threat model: the server that the e2ee is supposed to protect you from is the same one that serves you the code each time, which means that it would trivial to subvert it on-demand for targeted users with a negligibly-small chance of anyone catching them in the act. this is an example of cryptographic snake oil.
  • note that whether (or how often) they actually do subvert their encryption for targeted users is unproven; it has been alleged in court testimony that they do, and that the entire service is in fact a honey pot for doing exactly this (see my other comments for a link about that), but the testimony is perhaps not the most credible. however, the fact that they have the ability to do this (and without much chance of being caught) is undeniable and obvious to experts while being very much not obvious to most of their users.
• they are incessant blog spammers, writing advertorials about every vaguely-privacy-related news event to promote their products
• and, a new item on the list (not particularly surprising, given their history of shameless hucksterism): their README claims a clear policy against AI contributions but if you click that link and scroll up and click the three dots next to the commit message (“test plugin kit and nextcloud plugin”) you can see it says “Assisted-by: Claude:claude-sonnet-5”. (This kind of dishonest bandwagon-jumping-on is very on-brand for tuta.)

You can see some examples here

oh i see, that one doesn’t have a microphone but rather three microphones 🤯

(previously via hackaday i saw someone turn a disposable vape into a webserver but the use of microphones in them is news to me, thanks)