@starchturrets @justsoup @valpackett Well, the worst part is that I'm 100% sure LLMs were trained on stolen intellectual property... and either nobody noticed, or nobody cares?

I was working on a project, where we were the first ones to pair PCIe DSPs with x86 CPU. We ran into an unusual problem where client used two of those DSPs, and each one of them required 512+512+256MB of 32-bit BAR space (so it had to fit under 4GB).
One DSP was fine as default size for BAR space on Intel CPUs is 2GB, but it ran out of resources to allocate with both devices attached (which used 2.4GB on their own, plus PCH etc.).

It was a first for me, couldn't think of a solution from the top of my head and went to sleep. Client's engineers kept looking into it and when I woke up I had two links to... ChatGPT prompts in my mailbox. I thought "what a waste of time, but it's client so I guess I gotta read that".

First one was an absolute waste of time, in second one ChatGPT suggested that MmioSize pointer might increase bar spacing. I know Intel CPUs pretty well but this pointer isn't well explained anywhere - I checked FSP headers and it wasn't explained, in FSP Integration Guide you only see mention that if you allocate more memory for iGPU, you need to reduce MmioSize because of 4GB boundary limitation.

So I thought "weird but ok", I passed a pointer that set MmioSize to 2.8GB... and it worked. What the f...?
Surely most people would go "oh cool, it found a solution hooray thanks AI" but I went "how did the overclocked text predictor find that information? I refuse to use the code if I don't understand what it's doing.
Started digging - Intel documentation, headers, drivers and so on. Nothing explained it. Then I looked at FSP source code (which obviously is under NDA) where that pointer is being parsed... and I went "oh... shit".

There's no way that Intel uploaded their confidential code to train LLMs. I doubt any company leaked it either (because NDA is strict and penalties would destroy you). So the last explanation was that... OpenAI threw leaks (like ExConfidential) into their training dataset.
Any developer using those leaks would've been sued. I considered reporting this to Intel's legal team but then I went "hold on... what if OpenAI sues me for this or something along those ways? I can't afford the court case...".

I didn't end up reporting it, but if Intel's lawyers would get angry, it would get really ugly. That's the risk of using so-called "LLMs" for coding.

America's engine. Cartoon from one year ago.

#Trump #lies #democracy

@danirabbit Also people of other ethnicity in general.

But well said

If you’re upset and trying to guilt people of color for rightfully pointing out that nobody gave a shit until it happened to a white lady, maybe sit down and shut the fuck up

RIP human(-made) creativity? Cartoon for Dutch newspaper Trouw: https://www.trouw.nl/cartoons/tjeerd-royaards~bcb45712/

Note: this cartoon was made without any use of AI.

#AI #creativity #ArtificialIntelligence #BigTech

@pj @gruff how about non-EU folks?

It's been zero days since I lost time on a GH repo before realizing 'oh wait, this is slop, that's why none of this makes any sense' after 15 minutes.

Is there some nice community list/plugin that I could use to give me a big red border around repository pages of vibe coded garbeoleum?

I know about open-slopware, but that's not what I'm looking for - I want a baseline filter for obvious useless slop, not one where well established projects might fall into just because someone used an LLM once.

How hackers are fighting back against ICE when they are invading US cities and how to defend communities and hopefully protect them through clever use of technology https://www.eff.org/deeplinks/2026/01/how-hackers-are-fighting-back-against-ice

If Andrew "bunnie" Huang didn't exist, I'd swear he was a character out of a(n extraordinarily technologically well-informed) cyberpunk novel. Every time I interact with this legendary hardware hacker, he blows my mind with some project or insight that permanently alters how I think about tech.

-

If you'd like an essay-formatted version of this thread to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:

https://pluralistic.net/2026/01/09/quantity-break/#so-many-chips

1/

*slaps speaker* this thing can fit so much mainline in it

@ariadne @matrix please

hey @dpk can you do something about this?

https://codeberg.org/alpine

It is the same person who is impersonating Alpine on Telegram.

https://gitlab.alpinelinux.org/alpine/council/-/issues/699 for more details

edit: @dpk has done something about this. thanks for the swift response!

Your irregular reminder that Alpine Linux DOES NOT have any Telegram channels and any community use of the Alpine marks in a way which indicates an official relationship with or endorsement by the project is forbidden by our Code of Conduct.

Unfortunately, it is hard to get Telegram to do anything about this. We have been trying for years.

@fazalmajid @ariadne Okay, no worries :)

Yesterday evening I geeked on @postmarketOS and installed u-boot and a brand new version of postmarketOS on an "old" 2011 android 4.0 tablet. So happy to see this device with new up to date software 🤩🎉

#reuse #recycle #postmarketOS

@fazalmajid @ariadne Now debunking the other part: postmarketOS does support running Android apps thanks to Waydroid https://waydro.id/ (which runs an Android (LineageOS 20) lxc container on-top of your regular Linux) and Android Translation Layer https://gitlab.com/android_translation_layer/android_translation_layer (which is similar to wine, so it doesn't require an android container and can let you run apps directly). Other solutions exist in postmarketOS too. See the following wiki pages: https://wiki.postmarketos.org/wiki/Android_Translation_Layer https://wiki.postmarketos.org/wiki/Waydroid

@fazalmajid @ariadne FuriOS is not linux. It's Debian with a downstream Android kernel and an infinite amount of hacks (aka. Halium) to make that aging (typically 4.x) kernel work with debian userspace. It is closer to Android than it is to desktop Linux. And obviously it only works on their own phones (one of them just being a rebadged cheap Gigaset phone).
This means you're relying on an ancient Android kernel and lots of hacks that may break at any time, as well as a whole bunch more hacks to get all the android userspace blobs working, It may work now, but in no way is it sustainable. TBH I hate them calling it a "Linux phone" .. like I said it is much closer to Android than it is to Linux. Also, a lot of FuriLabs' Android emulation magic is proprietary and there is no reason why they cannot opensource/free it.

So, postmarketOS being less free than FuriOS? I'm sorry, where do you get this information from?

@davidgerard You're a year late.