Replying to
@inthehands thanks, AI summary, I never really thought of it that way before
Remote
technomancy
@technomancy@hey.hagelb.org
trying to catch the last train out of omelas
https://technomancy.us/colophon and my alt is @technomancy@icosahedron.website
how I run my instance: https://technomancy.us/201
1508 Followers
200 Following
50 Posts
Joined August 07, 2022
Replying to
@glyph@mastodon.social
ADULT: see the thing is it's really complicated
SECOND GRADER: is it, though?
Replying to
@datarama@hachyderm.io an "AI" model is most likely to "make you superfluous" not by doing tasks that you can do, but by crashing the worldwide economy so badly that no one can afford the luxury of paying a professional to develop custom software for them any more
Replying to
@whitequark@social.treehouse.systems not exactly "genAI-free" but the main reason I accepted my latest job is that I got an agreement that I would not be forced to use it myself (co-workers still do unfortunately, and if I got an offer from a place that banned it entirely, it would be hard to turn down)
Replying to
@glyph unfortunately the most likely outcome here is that management starts seeing that boilerplate goes from being a downside to a benefit, because it allows them to brag about LoC
Replying to
@aphyr@woof.group I'm shocked AWS reduced the bill
imagine going to the gas station and saying "I made a mistake; I meant to drive to Portland, Oregon and drove to Portland, Maine by accident; please can I get a refund on the fuel I spent?"
Replying to
Replying to
@eniko@mastodon.gamedev.place @baldur@toot.cafe I'm obviously very late in seeing what should have been obvious for years, but for me personally it has been awfully convenient to finally come to the conclusion that a better world can only happen after the end of US dominance right around the same time the US has made up its mind it is steadfastly determined to destroy itself
Replying to
@tael@yiff.life at least some of the time it's because bosses refused to allocate time for improving documentation for humans and now people are tricking into letting them invest in it because "it's for AI"
Replying to
@jonny@neuromatch.social rare case of reading far enough to see "oh, and it has an app!" and still wanting one anyway
Replying to
@be@floss.social @mhoye@cosocial.ca six entries currently on https://noai.starlightnet.work/list.html#:~:text=Languages (but four of them are lisps)
Replying to
@drahardja bahaha they didn't even bother to write a PR description; they just left it blank https://github.com/microsoft/vscode/pull/310226
Replying to
Replying to
@kentenmakto tried to put extratoenail.club in my shopping cart but it turns out their add to cart button doesn't actually work
Replying to
@Infrogmation my website: "oh no, it looks like you don't have an ad blocker installed! you better go fix that, please! I can wait, go do it now"
Replying to
@soapdog I have noticed that the react-based settings page is the part of gotosocial where I've noticed the most bugs, by a pretty decent margin
glad to see @liberachat@fosstodon.org clarifying their policy around LLMs in bots
while they're not banned altogether, channels that have LLM-powered bots are required to clearly disclose this to everyone joining
also really appreciated this angle:
Additionally, we’re going to codify the assumption that an LLM is capable of taking any action permitted by the bot/client whenever prompting occurs, regardless of what the prompt says. This reflects a growing corpus of real-world examples of AI assistants disobeying prompts and deleting important information. One of the practical consequences of this is that AI assistants, agents, and other LLM-driven autonomous clients will be effectively banned from Libera, as the assumption means we must also assume that they may spuriously join channels without permission.
potential career path for young folk entering the workforce:
blocklist sommelier
Replying to
@zyd no, when I tried it a year or two I ran into some pretty serious bugs that made it a non-starter for Fennel
I expect those have been fixed, but the timing wasn't right for me; I couldn't wait for the fixes before moving off git send-email
I would definitely be open to trying it, but I don't know how to start, because I don't want to make it an official thing for Fennel (adding more "blessed" patch submission methods would just be confusing at this point) and I don't have other projects that get patches
definitely a bit of a chicken/egg problem
Replying to
@krig@goto.liten.app yeah, on top of the democratic governance issue, I think it would make a huge difference if they made the login flow better for indie forgejos, in particular removing the need to manually register OAuth clients for every site you want to support login with: https://technomancy.us/#fn1
Replying to
@mia yeah, I guess I should have mentioned that I'd like this done somehow in a way that works with user.js since I've integrated that into my git dotfiles and I'd rather not add a new file and figure out how to integrate that into git, but a user stylesheet is probably the best way to do this
I just wish firefox weren't so resistant to dotfiles automation
looks like the codeberg resolution banning vibecoded projects passed! democracy in action!
I know being in software is just so exhausting right now, but I'm taking this as a reminder that we don't have to constantly be on the defensive and caught up into endless arguments
friends, if you're exhausted by arguing with boosters, please consider just stopping that completely once you realize what's happening and changing gears to work with people you've already identified as allies; there is so much work we still need to do if we are going to preserve "software that actually works and can be understood" in the coming years
https://codeberg.org/Codeberg/org/pulls/1253#issuecomment-19820434
Replying to
@aphyr@woof.group fun fact: pretty sure this features the only starfleet captain that's neither human nor vulcan that we will see for decades (he's Bolian)
Replying to
@hsza@social.tudbut.de ohhhh no no no
I had work I need to get done today ahhhh but you reminded me of 17776
=D
Replying to
a while back I asked about fiction written in unusual formats such as a spreadsheet or git commit history log, and yall delivered some pretty great examples
yesterday I found maybe the strangest format I've encountered yet: a work by @suricrasia@lethargic.talkative.fish told in lecture slides for a computer science class
https://suricrasia.online/unfiction/CSC218-Software-Precognance.pdf
so #emacs has a temporary ban on LLM contributions, but only while they wait for GNU to finalize a policy for the whole organization
myself and many other Emacs users don't have a lot of faith in GNU at this time; the ethically sound policy is very simple, and we feel that if they were going to the right thing, they would have done it by now
with that in mind, in the event that they do allow LLM code in Emacs, we will continue to use and develop a separate version of Emacs which is not contaminated
https://human-emacs.org
if you feel the same way, please join us and add your name to the list
Replying to
@mia yeah, reader-mode is a workaround for bad design and I'm glad to have it, but it would be better if a workaround weren't needed
tangentially related to the https://human-emacs.org project, I've been thinking about distro policies
several distros have no-LLM policies applying to packaging scripts, but afaict no one is tracking LLM infection in the upstreams themselves, probably because it's just a huge amount of work and no one distro could be expected to actually keep track of all that!
if I hadn't just started this Human Emacs project, I would be looking at coordinating across distros and coming up with some way to pool effort to track which upstreams are infected, which have strong protections in place, which are at risk, and how badly
similar to the "vibe-check" package metadata suggested by @cwebber@social.coop here: https://codeberg.org/guix/guix-consensus-documents/pulls/13#issuecomment-15774065
if you're looking at all this wondering "what can I do about all this?" I'd suggest approaching it from this direction, especially if you have experience packaging and working with distros
Replying to
@mhoye@cosocial.ca waiting on the edge of my seat here to see whether freebsd or openbsd will be the first to finalize their anti-LLM policy to help me make up my mind between the two
Replying to
@mcc@mastodon.social reading the article I was disappointed to find out they're only torturing the "open" models and not the ones from megacorps; weak
Replying to
@zyd@yap.zyd.lol I really wish sourcehut's own docs would lay off the "you must use git send-email!" propaganda; it is not good and if you insist on email you'll get better results by using attachments
I appreciated Tatham's take on that, and also the emphasis on "use whatever you as the contributor feel comfortable with"
Replying to
Replying to
@meena here's my attempt: https://git.sr.ht/~technomancy/fennel/tree/main/item/AGENTS.md
[no idea if it works because I've never used an LLM; and even if it works today who knows about tomorrow]
Replying to
@mattly@hachyderm.io "I had to enable parental controls to keep my dishwasher from attacking other computers" is one of those sentences which would leave me greatly amused in a PKD novel but I'm less fond of IRL
but it does make me pretty curious about what sites it's trying to hit; like I wonder if you could identify the command-and-control node it's getting instructions from, or a sample of the sites it's attacking assuming it's not encrypting the SNI in its outbound requests
Boosted by @joe@f.duriansoftware.com
Replying to
so the impression I get is that searx is like ... the GNU Social of free search; it's a precursor that represents an early shot at replacing the existing megacorp offering with something independent but it's not really usable as a serious replacement unless you are ideologically dedicated to avoiding the big bad one
https://docs.searxng.org/
I don't really want to use searx; I want to use the Mastodon of search--well, no, I want to use the Gotosocial of search--but the Mastodon of search doesn't exist yet and is going to be written by someone who's using the GNU Social of search, and the Gotosocial of search is going to be written by someone who's using the Mastodon of search
so I should probably at least poke around and get comfortable with it to see what they are up to
Replying to
@mattly@hachyderm.io yeah it's distressingly possible that your dishwasher company is a bright data customer and is pocketing a cut of the proceeds they get from deliberately deploying malware into your house, even without any vibecoding to blame
Replying to
@aphyr it's a shame we'll never know how much of it is due to load and how much is due to getting high on their own supply
made a blog post about code hosting setups and what your options are for how to #getOffGithub and how it worked out for me https://technomancy.us/204
I love that codeberg works toward making it easy to start your own alternate-codeberg with forgejo; they even list out a bunch of recommended alternate instances in their "what even is codeberg" page: https://docs.codeberg.org/getting-started/what-is-codeberg/#what-is-codeberg-e.v.%3F
but I wonder if there's an opportunity to take it beyond the technical; forgejo is only half of what makes codeberg interesting, the other half is the democratic governance
I think in the interests of avoiding having all our eggs in one basket, it would be really cool to see a "how to start your own democratically-governed codeberg-like organization" tutorial
we need more of these!