Hrm. I… hrm.

[mcc@mastodon.social]

Hrm. I… hrm. I… I'd consider using this or basing things on this. Hrm.

https://social.kernel.org/objects/f9cc15c0-fb3b-40e9-b561-049d829872db

No COPILOT.MD, which is not proof of human-authored software, but for an MS product seems like a strong hint toward it.

[dysfun@social.treehouse.systems]

@mcc it's microsoft research, it might not have had copilot shoved on them yet.

[mcc@mastodon.social]

Been wanting to try an OS project and of course, in Rust, why not. The two things I've been considering starting with are the rust "uefi" crate, or Redox OS. Rust-uefi, Redox, and Litebox I'd of course be targeting for different reasons, but if the meta-goal is to learn "OS things", the proximate reasons aren't so important.

[dcbaok@fosstodon.org]

@mcc I've been meaning to check out tock os, a small os for embedded, in rust. https://www.tockos.org/

[sarahjamielewis@mastodon.social]

@mcc https://github.com/microsoft/litebox/blob/main/.github/copilot-instructions.md / https://github.com/microsoft/litebox/commit/5fdd6af324102c24aea277507bc86249a3d624f1

[mcc@mastodon.social]

@sarahjamielewis ah… i see. thank you.

well, maybe i try to mod redox into doing the things litebox does by hand.

[mcc@mastodon.social]

Hmm, nope, copilot.md there just more insidious.

https://mastodon.social/@sarahjamielewis/116008743274901543

What I want is to be able to run a full-computer software stack with no "AI"-gen code. Unless/until we get mandatory product-labeling laws, this is incompatible with using any open-source code outside my control, as fake-code distributors often hide it. So I find myself drawn to something like "retrocomputing, but not old": Can I make a minimal full software stack I *wrote myself*. Strip the "OS" until it's in my grasp

[mcc@mastodon.social]

When I suggest projects should introduce rules against randomly-generated ("AI") code, the primary (only?) argument against such a rule is "if you try, we'll just do it in secret". Less an argument than a taunt. Okay, you've convinced me, fine: Once "AI" exists, society no longer exists. I can only trust myself. I'll write the entire OS myself if that's what it takes to not touch anything Copilot/Claude touched. Try to sprint from UEFI to a wasm interpreter, run all third party stuff in there.

[mcc@mastodon.social]

This might take twenty years, but I've got those.

[mcc@mastodon.social]

So, with the above as context, Microsoft Litebox is *very* interesting. You could very well write an OS yourself, but writing a stack of *all* apps you use is harder. Hundred Rabbits did it! But what I probs want is the minimal basis for some *existing standard*, webtech or POSIX. Microsoft here is creating "a minimal basis for POSIX", but for a different reason: Instead of "I want it small so I can write it myself" it's "I want it small so it's good for Docker/Hyper-V".

… but MS means Copilot.

[bob@feed.hella.cheap]

@mcc minix3 exists

[dalias@hachyderm.io]

@mcc The "we'll just do it in secret" is a bluff we can call. The AI slop shitheads aren't talented enough to do it in secret, and the statement of values ensures you'll have a community that will catch and reject their bad attempts. The reason they say that is not even that they have the will to try to do it in secret. It's that they want us to believe their sorry world is inevitable. They want us to give up.

[snaums@toot.kif.rocks]

@mcc but isn't that just unikernels? I mean, unikraft exists, they have a docker-like description file, can strap a unikernel or linux under the application and run it in a VM. Ok, optimized for boot time, not security.

[dysfun@social.treehouse.systems]

@mcc "sure is a mystery i don't want to accept AI code with AI coders like you, huh?"

[kevingranade@mastodon.gamedev.place]

@mcc I have no intention of arguing against your main conclusion, but there's an aditional flaw in that LLM supporters tend to think that their "secret" contributions are indistinguishabe from a regular code contributions, which couldn't be further from true.
In practice they're incredibly tone deaf and overly-aligned with LLM causes in the first place.

[mcc@mastodon.social]

@snaums I am not a member of the Litebox team and cannot speculate why they did not take this path. Possibilities include:

- The proposal you describe is already being done, and therefore not an interesting project for a research team.

- Possibly some important Linux applications necessarily involve a web of processes with IPC, a poor fit for unikernel.

- Possibly some single-process applications "should" be untrusted due to their input (for example: a web browser that executes sandboxed JS)

[mcc@mastodon.social]

@bob Interesting

[nina_kali_nina@tech.lgbt]

@mcc sometimes I too want to go permacomputing route. Maybe DuskOS approach could be a good one. Maybe using pen and paper is even better.

[klara@wandering.shop]

@snaums @mcc My understanding, which comes from someone who's hacked a lot on MirageOS, is that they're great and fun for small applications, but often you want a big piece of common functionality that the libraries your unikernel framework provides don't have, and, are you up for building it?

Which is a LOT of (design+implementation) work. The people with tons of money aren't motivated to fund this work. But you can't reuse existing software from outside that unikernel project. So it's on you.

[emaytch@mastodon.social]

@mcc i’m always so glad to learn that there’s someone else with my basic attitude towards life and doing things out of spite