self

in the same vein as AGENTS.md, CLAUDE.md, and a fuckton of other repo spam, I present AAA-NO-SLOP.md, a file for humans viewing repositories that signals two things:

• this repository doesn’t accept LLM contributions of any kind
• every other LLM instruction in this repository (AGENTS.md, CLAUDE.md, and all the rest) is poisoned and designed to deter LLM use

enjoy!

for any guests who stumble upon this thread: no I’m not entertaining discussion on why I’m doing this or how I shouldn’t do it

our version of lemmy is old enough that clients like mlem are starting to break due to API drift, so I’m finally upgrading us to the latest stable version of lemmy. this will involve a bit of downtime and potentially a number of breakages; keep an eye out for anything that doesn’t look right after the upgrade and let us know!

after some extended downtime, I rolled out the following changes to our instance:

• pict-rs was migrated to version 0.4 then 0.5. this should hopefully fix an issue where pict-rs kept leaking TCP sockets and exhausting its resources, leading to our image uploads and downloads becoming non-functional. let me know if you run into any issues along those lines!
• NixOS was updated to 24.11.
• the instance's storage was expanded by 100GB. this increased the monthly bill for our instance by €1.78 per month. to keep the bill low, I disabled an automated backup feature that became unnecessary when we started doing Restic backups.

I have one more thing I want to implement before our big Lemmy upgrade; I expect I should be able to fit it in tomorrow. I'll update this thread with details when I start on it.

since we’ve been experiencing a few image cache breakages, I’m scheduling some maintenance for January 24th at 8AM GMT to upgrade our pict-rs version, increase the total amount of storage available to our production instance, and do a handful of other maintenance tasks. this won’t include a lemmy upgrade, but I plan to do one soon after this maintenance round. I anticipate the maintenance should take around 2-4 hours, but will post updates on the instance downtime page and Mastodon if anything changes.

we have a WriteFreely instance now! I wrote up a guide to why it exists, why it's so fucking janky, and what we can do to fix it.

this is somewhat of a bigger update, and it's the product of a few things that have been in progress for a while:

email

email should be working again as of a couple months ago. good news: our old provider was, ahem, mildly inflating our usage to get us off their free plan, so this part of our infrastructure is going to cost a lot less than anticipated.

backups

we now have a restic-based system for distributed backups, thanks to a solid recommendation from @froztbyte@awful.systems. this will make us a lot more resilient to the possibility of having our host evaporate out from under us, and make other disaster scenarios much less lethal.

writefreely

I used some of the spare capacity on our staging instance to spin up a new WriteFreely instance where we can post long-form articles and other stuff that's more suitable for a blog. post your gibberish at gibberish.awful.systems! contact me if you'd like an invite link; WriteFreely instances are particularly vulnerable to being turned into platforms for spam and nothing else, so we're keeping this small-scale for instance regulars for now.

alongside all the ordinary WriteFreely stuff (partial federation, a ton of jank), our instance has a special feature: if you have an account, you can make a PR on this repository and once it's merged, gibberish will automatically pull its frontend files from that repo and redeploy WriteFreely. currently this is only for the frontend, but there's a lot you can do with that -- check out the templates, pages, less, and static directories on the repo to see what gets pulled. check it out if you see some jank you want to fix! (also it's the only way to get WriteFreely to host images as part of a post, no I'm not kidding)

what's next?

next up, I plan to turn off Hetzner's backups for awful.systems and use that budget to expand the node's storage by 100GB, which should increase the monthly bill by around 2.50 euros. I want to go this route to expand our instance's storage instead of using an object store like S3 or B2 because using block storage makes us more resilient to Hetzner or Backblaze evaporating or ending our service, and because it's relatively easy to undo this decision if it proves not to scale, but very hard to go from using object storage back to generic block storage.

after that, it'll be about time to carefully upgrade to the current version of Lemmy, and to get our fork (Philthy) in a better state for contributions.

as always, see our infrastructure deployment flake for more documentation and details on how all of the above works.

this post has been making the rounds on Mastodon, for good reason. it’s nominally a post about the governance and community around C++, but (without spoiling too much) it’s written as a journey packed with cathartic sneers at a number of topics and people we’ve covered here before. as a quick preview, tell me this isn’t relatable:

This is not a feel good post, and to even call it a rant would be dismissive of the absolute unending fury I am currently living through as 8+ years of absolute fucking horseshit in the C++ space comes to fruition, and if I don’t write this all as one entire post, I’m going to physically fucking explode.

fucking masterful

an important moderator note for anyone who comes here looking to tone police in the spirit of the Tech Industry Blog Social Compact: lol

this article is about how and why four of the world’s largest corporations are intentionally centralizing the internet and selling us horseshit. it’s a fun and depressing read about crypto, the metaverse, AI, and the pattern of behavior that led to all of those being pushed in spite of their utter worthlessness. here’s some pull quotes:

Web 3.0 probably won’t involve the blockchain or NFTs in any meaningful way. We all may or may not one day join the metaverse and wear clunky goggles on our faces for the rest of our lives. And it feels increasingly unlikely that our graphic designers, artists, and illustrators will suddenly change their job titles to "prompt artist” anytime soon.

I can’t stress this point enough. The reason why GAMM and all its little digirati minions on social media are pushing things like crypto, then the blockchain, and now virtual reality and artificial intelligence is because those technologies require a metric fuckton of computing power to operate. That fact may be devastating for the earth, indeed it is for our mental health, but it’s wonderful news for the four storefronts selling all the juice.

The presumptive beneficiaries of this new land of milk and honey are so drunk with speculative power that they'll promise us anything to win our hearts and minds. That anything includes magical virtual reality universes and robots with human-like intelligence. It's the same faux-passionate anything that proclaimed crypto as the savior of the marginalized. The utter bullshit anything that would have us believe that the meek shall inherit the earth, and the powerful won't do anything to stop it.

we’ve exceeded the usage tier for our email sending API today (and they kindly didn’t email me to tell me that was the case until we were 300% over), so email notifications might be a bit spotty/non-working for a little bit. I’m working on figuring out what we should migrate to — I’m leaning towards AWS SES as by far the cheapest option, though I’m no Amazon fan and I’m open to other options as long as they’ve got an option to send with SMTP

after the predictable failure of the Rabbit R1, it feels like we’ve heard relatively nothing about the Humane AI Pin, which released first but was rapidly overshadowed by the R1’s shittiness. as it turns out, the reason why we haven’t heard much about the Humane AI pin is because it’s fucked:

Between May and August, more AI Pins were returned than purchased, according to internal sales data obtained by The Verge. By June, only around 8,000 units hadn’t been returned, a source with direct knowledge of sales and return data told me. As of today, the number of units still in customer hands had fallen closer to 7,000, a source with direct knowledge said.

it’s fucked in ways you might not have seen coming, but Humane should have:

Once a Humane Pin is returned, the company has no way to refurbish it, sources with knowledge of the return process confirmed. The Pin becomes e-waste, and Humane doesn’t have the opportunity to reclaim the revenue by selling it again. The core issue is that there is a T-Mobile limitation that makes it impossible (for now) for Humane to reassign a Pin to a new user once it’s been assigned to someone.

as I was reading through this one, the quotes I wanted to pull kept growing in size until it was just the whole article, so fuck it, this one’s pretty damning

here’s a thin sample of what you can expect, but it gets much worse from here:

Internal conversations at Nvidia viewed by 404 Media show when employees working on the project raised questions about potential legal issues surrounding the use of datasets compiled by academics for research purposes and YouTube videos, managers told them they had clearance to use that content from the highest levels of the company.

A former Nvidia employee, whom 404 Media granted anonymity to speak about internal Nvidia processes, said that employees were asked to scrape videos from Netflix, YouTube, and other sources to train an AI model for Nvidia’s Omniverse 3D world generator, self-driving car systems, and “digital human” products. The project, internally named Cosmos (but different from the company’s existing Cosmos deep learning product), has not yet been released to the public.