f_ 🇵🇸
Posts
393Following
93Followers
170Don't PM here; PMs are disabled. Instead, reach out via email or IRC/XMPP.
Followers maybe manually approved, please have a bio. More likely to follow-back if we met somewhere else (e.g. IRC).
Note:
*** Fascism, racism, as well as any other kind of bigotry NOT TOLERATED!
*** MDNI/18+ accounts also NOT TOLERATED!
*** You are entering a fascism-FREE zone.
*** Opinions are my own and do not represent those of other projects
*** I contribute to. I try to not boost images without alt-text. Please
*** use alt-texts to describe your images, so that blind folks also
*** get it. Thanks! ^^ (if your image does not contain alt text and I
*** boost it anyway, I may add alt-text as a reply, you are encouraged
*** to add it to the image directly or improve upon it)
#nobot
f_ 🇵🇸
fun
@GrapheneOS @domi @elly @mei @ariadne
I'm glad you're talking to @ariadne about this. She's nice.
I hope both projects can collaborate on things one day, I think that would be very beneficial tech-wise, as you (GrapheneOS team) have a lot of insights on security and such for example, so we can share notes etc... :)
Take care of yourself!
I'm glad you're talking to @ariadne about this. She's nice.
I hope both projects can collaborate on things one day, I think that would be very beneficial tech-wise, as you (GrapheneOS team) have a lot of insights on security and such for example, so we can share notes etc... :)
Take care of yourself!
1
0
0
f_ 🇵🇸
repeated
repeated
f_ 🇵🇸
repeated
repeated
Sven Peter
sven@social.treehouse.systemsFor those who missed my #Asahilinux #39c3 talk, it's available at https://media.ccc.de/v/39c3-asahi-linux-porting-linux-to-apple-silicon now.
I've also just pushed my slides to https://github.com/svenpeter42/39c3 and uploaded them as PDF to https://cfp.cccv.de/39c3/talk/YGHB9K/
10
6
0
f_ 🇵🇸
fun
@zyx @Logical_Error @cas apparently they tried to 'fund' GrapheneOS then apparently blackmailed them and it didn't go well.
1
0
0
f_ 🇵🇸
fun
@Logical_Error @cas The KF part I just saw this in a fediverse thread
In any case FUTO is quite a questionable orga: https://drewdevault.com/2025/10/22/2025-10-22-Whats-up-with-FUTO.html
In any case FUTO is quite a questionable orga: https://drewdevault.com/2025/10/22/2025-10-22-Whats-up-with-FUTO.html
1
0
2
f_ 🇵🇸
fun
@Logical_Error @cas Louis Rossmann is apparently a KiwiFarms lurker and has harassed GOS in the past with his FUTO project thing which also platforms fascists.
Would not recommend watching
Would not recommend watching
4
0
5
f_ 🇵🇸
repeated
repeated
justsoup 💛 musl
justsoup@mstdn.social
Harassment, AI
Show content
For those not following, open-slopware, the repository dedicated to cataloging projects that use LLMs as development, was taken down today after the main organizer was harassed by Bluesky and Reddit pro-ai individuals. As such, development has been pretty fragmented today, but it has reconsolidated at https://codeberg.org/ai-alternatives/llm-afflicted-software.
No matter whether you like LLMs or not, I don't think harassment is ever the answer... for anything...
1
4
0
@mei @GrapheneOS @domi @elly note that I have PMs disabled on fediverse, if you want to share these in PM please do so to "f_" on irc.oftc.net or @funderscore:postmarketos.org
1
0
2
f_ 🇵🇸
fun
@mei @domi @elly @GrapheneOS As far as I know we have not received anything like that so far.
I am also not aware of any team member actively raiding GrapheneOS chats. If you have proof it would be nice to link to them because so far it's mostly been "trust me bro" (no offense intended).
Take care
I am also not aware of any team member actively raiding GrapheneOS chats. If you have proof it would be nice to link to them because so far it's mostly been "trust me bro" (no offense intended).
Take care
2
0
2
f_ 🇵🇸
repeated
repeated
@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.
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.
1
1
0
f_ 🇵🇸
repeated
repeated
