Egregoros

Signal feed

Mary Leigh

@lw@mastodon.bsd.cafe

unfortunately, computers

Posts

Latest notes

looking through my MikroTik support tickets, i see that i reported this in 2024 as SUP-168353 and the response was:

"We are aware of the issue, hopefully it will be fixed in the future."

i took this to mean "we might fix it in the future" but i guess they were just hoping it would fix itself, or something.

#mikrotik

trying to work out what to do with an arm64 Mac Mini...

- macOS is useless as a server (e.g., it overwrites all your /etc changes on every upgrade)
- FreeBSD doesn't work since we don't have NVMe support
- OpenBSD doesn't work because aq(4) only supports firmware version 2, and this has version 1
- the only supported Linux distribution seems to be Fedora, which i don't want to use, and that doesn't even support installing; you have to "install" it from macOS, which does not sound reliable

any other ideas?

@elliot i'm not sure what the right scale for "home network complexity" is but i guess... fairly?

it has externally 2x IPv4 /28s and a IPv6 /48 that are subnetted and need to be routed to the right place, some other internal-only networks (RFC1918/ULA), and is part of a VPN covering three sites including this one

you know, everyone performatively avoiding any software with a single line of LLM-generated code in it could spend their time doing something much more productive, such as burning down AI datacentres.

one of these achieves your aims, the other does not. just saying.

"EPOW_POWER_OFF 7 The system will lose power. The hardware ensures that at least 4 milliseconds of power within operational thresholds is available after signalling an interrupt."

so, you're an operating system kernel, and you've just been informed you only have 4ms to live. what do you do?

- sync disks: usually takes longer than 4ms, and an interrupted sync on UFS can do more damage than not syncing

- log the event: but what is the point of the log if it's not committed to disk?

- try a normal immediate shutdown and hope we actually have more than 4ms left?

i see people fleeing GitHub for Codeberg, which is understandable, but Codeberg is only one good DDoS from being down for a week.

the fix for this isn't to move from one provider to another, it's for people to host their own projects, or ask a friend to host it. Git has a built-in email workflow that works great, you do not need "pull requests" to accept contributions.

@feld the difference is Amazon was based on a well-known business model: undercut competitors through economy of scale. that would not have been a risky bet.

there's as yet no "economy of scale" in AI, only scaling at the same cost-per-user to support more users.

perhaps later models will change that, but perhaps not; we really don't know right now. i would certainly not be buying AI stocks right now, if i was rich enough to own stocks.

@feld it's not clear to me that "we can generate whatever we want". in the sense that i could go on ChatGPT right now and ask for "an image of a woman fretting over how ridiculous capitalism is", yes, but OpenAI is losing billions of $ every year to let me do that.

no one has yet demonstrated that LLMs are sustainable either financially or environmentally.

@feld i can appreciate that, for example, without copyright it would be a lot more difficult for people to make money from things like stock photography and artwork, because it's the threat of copyright litigation (rather than actual litigation) that encourages companies to license that rather than just stealing it.

but in a general sense, yes, the entire system is broken and really only benefits middlemen and legacy copyright owners (Disney, etc).