> We are completely overwhelmed, of course.

[hipsterelectron@circumstances.run]

@daemonspudguy @zzt @lykso @c0dec0dec0de is there more information about this anywhere?

[daemonspudguy@app.wafrn.net]

@hipsterelectron@circumstances.run @zzt@mas.to @lykso@tiny.tilde.website @c0dec0dec0de@hachyderm.io

The kernel it's using is Linux-libre (already a bad sign anyways because that kernel can't update microcode at all) version 5.10, which is absolutely ancient.

[daemonspudguy@app.wafrn.net]

@hipsterelectron@circumstances.run @zzt@mas.to @lykso@tiny.tilde.website @c0dec0dec0de@hachyderm.io

I would try to find a specific source about their stance itself but their website is utterly impenetrable because it's all philosophical wankery that I can't make sense of because my eyes physically can't read it. The kernel version, though, is just right on the package list. Last updated August 2025.

[daemonspudguy@app.wafrn.net]

@hipsterelectron@circumstances.run @zzt@mas.to @lykso@tiny.tilde.website @c0dec0dec0de@hachyderm.io

Oh my god I found one. It doesn't mention the kernel specifically but the fact that the kernel is clearly included in it makes it clear that that's their position.
https://wiki.hyperbola.info/doku.php?id=en:philosophy:reject_machinelearning#outcome_and_decisions_for_hyperbola

#outcome_and_decisions_for_hyperbola

[ophiocephalic@kolektiva.social]

@zzt
Identity scans for everything on the internet. Thank you Meta!

[pastelfluffyfox@meow.social]

@zzt feeling like gandalf when he read that journal about the fall of moria, only it's about LLMs coming to the linux kernel

[daemonspudguy@app.wafrn.net]

@hipsterelectron@circumstances.run @zzt@mas.to @lykso@tiny.tilde.website @c0dec0dec0de@hachyderm.io

Your list of operating systems ain't exactly large if LLM extruded code is a complete dealbreaker. Especially if you're exposing something on the internet.

[daemonspudguy@app.wafrn.net]

@hipsterelectron@circumstances.run @zzt@mas.to @lykso@tiny.tilde.website @c0dec0dec0de@hachyderm.io

@mrmasterkeyboard mentioned.

[lykso@tiny.tilde.website]

@daemonspudguy @hipsterelectron @zzt @c0dec0dec0de Yeah. Longer-term plan is developing something on top of seL4 with (probably) Idris 2, but I need something for hosting in the meantime.

I'm also planning to do this in a private-ish environment (hence the tilde), releasing everything under a license that precludes use by any entity that is not worker owned and controlled, and restricting access to the source code along the same lines.

[cap_ybarra@beige.party]

@dysfun @zzt it's evals eval'ing evals, and evals eval'ing those evals, and surely that's enough evals to not have to pay the nerds anymore right?

[hipsterelectron@circumstances.run]

@daemonspudguy @zzt @lykso @c0dec0dec0de that's why i'm making my own kernel

[hipsterelectron@circumstances.run]

@daemonspudguy @zzt @lykso @c0dec0dec0de operating systems like linux are extraordinarily inefficient

[c0dec0dec0de@hachyderm.io]

@lykso @hipsterelectron @daemonspudguy @zzt FUCK, tmux, is nothing sacred?!

[daemonspudguy@app.wafrn.net]

@hipsterelectron@circumstances.run @zzt@mas.to @lykso@tiny.tilde.website @c0dec0dec0de@hachyderm.io

I'm curious, is it a microkernel?

[daemonspudguy@app.wafrn.net]

@hipsterelectron@circumstances.run @zzt@mas.to @lykso@tiny.tilde.website @c0dec0dec0de@hachyderm.io

No. Sacredness was killed in the year 2020 along with all of humanity. 2020 was actually the apocalypse and this is actually a purgatory for all of humanity. /j

[hipsterelectron@circumstances.run]

@daemonspudguy @zzt @lykso @c0dec0dec0de yes but not in the L4 style. a "monolithic" kernel actively abdicates security boundaries or protocols of any sort and is impossible to optimize to application-specific access patterns. ring 0 is for memory allocations and not for hardware. explicit message buffering (unlike seL4) enables strong guarantees around data persistence and data copies can be used to safely transfer across security contexts as well as memory hierarchy regimes. the bullshit linux does to pass around pages to avoid copies is incredibly archaic and it's very cache-unfriendly

[hipsterelectron@circumstances.run]

@daemonspudguy @zzt @lykso @c0dec0dec0de most of POSIX can be supported but not fucking pthreads. the entire concept of a mutex is completely broken and that's why deadlocks are even allowed to happen. haven't figured all that out yet but i think any constructs built on top of c11 atomics need to have clear semantics for progress. the ring buffer i've built supports stuff like "block until N bytes are available to write" and it makes parallelism so much easier when data structures provide guarantees around data flow rather than control flow

[hipsterelectron@circumstances.run]

@daemonspudguy @zzt @lykso @c0dec0dec0de alexia massalin did the absolute coolest fucking shit ever with only an atomic compare-and-swap instruction in her synthesis kernel https://pages.cs.wisc.edu/~bart/736/papers/synthesis.pdf i'm doing a lot more with filesystem i/o and scheduler control because the scheduler is a security boundary and no process should have a global view of all files

[lykso@tiny.tilde.website]

@hipsterelectron That sounds fantastic, TBH. Really looking forward to seeing what you end up writing!
@daemonspudguy @zzt @c0dec0dec0de

[hsza@social.tudbut.de]

@zzt “we are drowning in slop, the solution is obviously more slop”