the initra mf
@me@doasu.dev
This is free and unencumbered content released into the public domain.:3 () { :3 | :3 & }; :3 >:3 # >:3c
82 following, 23 followers
NepoRC 2.13.7 is starting up enbyOS
* Mounting gender filesystem ...
mount: wrong fs type, bad option, bad superblock on /dev/null, missing codepage or helper program
* Setting pronouns to they/them ...
* Setting timezone to UTC+2 ...
Starting about-me runlevel
* Greeting user ...
Welcome to my page! (doasu.dev)snac login: me
Password:
Last login: this week (localhost)
~ % _
Changed default: for newly created instances, disable_inbox_collection is set to true (see snac(8) for more information). The reason is because it seems to be used for harrasing people.
Changed default: for newly created instances, disable_history is set to true (see snac(8) for more information). The reason is because archived history files don't reflect reality after posts are deleted or modified (they always have been an ugly kludge).
Changed default: in previous versions, posts with a scope of unlisted were shown in public pages and RSS feeds. Now, they are no longer shown. If you want to get back to previous behaviour, use a new toggle in the User Settings section (see snac(1) for more information).
New admin configuration option: if the purge_static value is set to true in server.json, each user's static directory is explored and those files there that are no longer attached to any post or referenced anywhere are deleted. See snac(8) for more information about those cases where you may not want to enable this option.
Allow serving files from subdirectories of the static/ subdirectory (contributed by la_ninpre).
Minor tweak to webfinger code to handle Hubzilla's peculiarities.
Fixed a search case where URLs to GotoSocial statuses were misidentified as accounts.
Accounts that follow you are now marked with a thumb-up emoji, because followers are adorable people.
Fixed some account export errors.
Fixed an incorrect hash in post links.
Show an account's location link in the people page, if they have one.
Mastodon API: Fixed hashtags loosing the link after editing a post, minor tweak in access token processing (contributed by trondd555).
Drop usage of PATH_MAX (contributed by sergiodj).
New Polish translation (contributed by kpm).
Updated German and Czech translations (contributed by zen and pmjv).
If you find #snac useful, please consider buying grunfink a coffee or contributing via LiberaPay.
> Allow serving files from subdirectories of the static/ subdirectory (contributed by la_ninpre)
@grunfink so I can finally host emoji icons on same domain?
as the world turns, does does single-binary #fediverse frontends. just implemented background pollling for notifications with a subtle indicator. still sitting at ~14mb binary tested against #pleroma #mastodon #gotosocial #mitra
@jae 👀 I srsly wonder what this software is…
@mstiemisty it's a standalone front end for the fediverse written in go. let's you run a small 14mb binary to interact with the network.
going to bring my infra online today forging.on.spiteful.systems coming soon
Microsoft reports AI is more expensive than paying human employees
Link: https://fortune.com/2026/05/22/microsoft-ai-cost-problem-tokens-agents/
Discussion: https://news.ycombinator.com/item?id=48244434
LOL
The reports may throw cold water on the bets tech’s biggest firms have placed on the technology. While some cling to the promise of an AI “renaissance” or “revolution,” the cost of adoption is proving a stubborn bottleneck. These developments also suggest that the economics of replacing or augmenting human labor with AI may be more complicated than some early forecasts originally implied. That echoes what Bryan Catanzaro, vice president of applied deep learning at Nvidia, recently said in an interview with Axios.
“For my team, the cost of compute is far beyond the costs of the employees,” he said.
https://fortune.com/2026/05/22/microsoft-ai-cost-problem-tokens-agents/
Just in time for #ohs2026 the research group I'm working at published our design, including the firmware I've been working on for the past year. Come by and say hi, I'm on shift duty at our assembly right now :3
codeberg.org/beecon
@hermlon what is Beecon? I can't really tell from the READMEs
@esoterra that's a good point, we should add a general section about the project somewhere there. For now, here's a poster, and there's also some (slightly outdated) information on the page of our chair www.tu.berlin/en/raumfahrttechnik/research/current-projects/beecon
It is my 3rd month of running #snac #snac2 instance. I wanted to share a few words about how awesome this software is.
Few points that make snac icredibly good for my usecase are:
After 3 month of posting (2 active users and 3 semi active users) my data dir is ~500MB and memory usage is ~200MB
it means i can keep running it on my infrastructure without even thinking too much about load
@grunfink@comam.es thanks for such a awesome piece of software 🩷🩷🩷
After months of testing, weeks of writing, here is finally my new full Linux install guide: "Interim Install Guide: KDE Neon User Edition for a professional digital painter workstation": https://www.davidrevoy.com/article1145/interim-install-guide-kde-neon-user-edition-for-a-professional-digital-painter-workstation
Everything about Drawing Tablet, Color Management and Calibration/Profiling on KDE Plasma 6 Wayland, and all to survive the packaging mess with eg. built-in tutorial to revert version of Flatpak, scripts to beta-test Krita, and more!
The most complete article I ever published!
Quick mirror of my article on my other server: https://www.peppercarrot.com/extras/html/2026-05-23_KDE-Neon-for-digital-painter-guide_mirror/ because my blog is in a overload situation, too many visitors, I hope it helps!
La tablette de plainte à Ea-nasir, écrite en cunéiforme il y a près de 3 800 ans en Mésopotamie.
Un client nommé Nanni y reproche au marchand Ea-nasir de lui avoir vendu du cuivre de mauvaise qualité et de l’avoir traité avec une délicatesse commerciale très discutable.
Bref, le plus vieux avis client mécontent connu ressemble déjà à : “produit non conforme, vendeur peu aimable, je déconseille.” Certaines traditions traversent vraiment les siècles.
Via @2tout2rien mais avec #Alt !
In around two weeks I’m migrating from iOS to a postmarketOS phone. Please leave app suggestions, I’m definitely going to need them!
Waydroid is… fine, but not preferred Please avoid android apps unless it’s something that is like confirmed to work well on waydroid, is proprietary or something so has no alternatives or something like that. I’m not launching waydroid to listen to music, to read my e-mails or to use xmpp.
Some examples of what I’m looking for, probably for Plasma Mobile:
And also just anything y’all find useful in your daily life with a postmarketOS daily driver, I’m sure I’m forgetting a lot of stuff here.
I know some of those exist for desktop Linux, but I’m obviously looking for stuff made with phones in mind (so like… apps made for gnome should be automatically fine maybe?)
(Also, has anyone managed to get nix’s system-manager to work on postmarketOS? There’s no way I’m configuring all of this by hand every time I need to flash it lmao)
boostedIf you run a peertube instance and have not patched in the past 4 hours, you are way behind and likely have been compromised. The latest patch will help clean up the mess.
See here: https://github.com/Chocobozzz/PeerTube/releases/tag/v8.1.8
To make the "zfs list" output more script-friendly, you can suppress the
output of the headers for each column by passing the -H parameter:
zfs list -H
Another helpful option for script writers is -p, which displays the numbers
in non-rounded, exact values:
zfs list -p
-- Benedict Reuschling <bcr@FreeBSD.org>
I made a thing! Or rather, ported a thing? It's oneko, but usable on a KDE Wayland session. That's basically it!
https://codeberg.org/snowkat/koneko
https://git.2ki.xyz/snow/koneko
It's not a 1:1 port of oneko-sakura, but I wanted to replicate the logic and customization where possible.
There are some features I still want to add, and I'm sure there's some bugs (please report them on the repo if you find them!). But if you want oneko chasing your cursor around your desktop and you use KDE, it does that!
info on the github breach appears to only be available on xitter 🙄 , I fished it out for you.
gonna gently push back that there's no reason (according to github's version of the story) to associate this with AI or with spectacular incompetence on the part of the employee; the issue is that industry standard, extremely widely used text editor Visual Studio Code has a big button that says "click here to add useful functionality to do your job" that has a 1% chance of installing ransomware
@0xabad1dea I'm honestly not sure if you're joking or if this is literally true.
@Nephrite @0xabad1dea 1% is maybe a bit exaggerated but VS Code marketplace is kinda notorious for malware
@ratsnakegames @0xabad1dea That sounds pretty bad. Don't they do reviews or anything?
@Nephrite @ratsnakegames @0xabad1dea The problem IMO is a complete lack of sandboxing. You can have completely legitimate extension one day, then next day it gets updated with a compromised version (perhaps via a dependency)
@radex @Nephrite @0xabad1dea you cannot meaningfully sandbox an extension whose functionality includes compiling and running code from the user's workspace
@ratsnakegames @Nephrite @0xabad1dea Sure. I'd rather say that _not every_ extension can be meaningfully sandboxed.
Required permissions could be clearly displayed and those that require full unsandboxed access could be additionally flagged.
IMO this would go a long way towards reducing risk of pwnage via extensions. Long process, sure, but worth it.
IFIN - The Independent Federated Intelligence Network » 🌐
@ifin@infosec.exchange
We regret to inform you that yet another GitHub attack is underway—this time compromising GitHub Actions with infostealer scripts.
https://discourse.ifin.network/t/5600-github-accounts-compromised-in-megalodon-attack/490
Out: robots.txt
In:<IfModule mod_rewrite.c>
RewriteEngine On
RewriteCond %{HTTP_USER_AGENT} ^(Googlebot|Googlebot-Image|Googlebot-Video|Googlebot-News|Storebot-Google|Google-InspectionTool|GoogleOther|GoogleOther-Image|GoogleOther-Video|Google-CloudVertexBot|Google-Extended|APIs-Google|AdsBot-Google-Mobile|AdsBot-Google|Mediapartners-Google).*$ [NC]
RewriteRule ^(.*)$ http://goatse.cx [L,R=301]
</IfModule>
@katzenberger @cR0w I've been serving a 32 GB HTML file pre-compressed with zstd to 2 MB to some Meta scrapper for the last two weeks.
a terrifying interaction at the ongologist: I ask a question, he opens up chatgpt and asks the same question to it, in front of me. Lets the phone talk back and then repeats what the phone said.
bro i could've done that myself. what are you here for
If you were to guess, which of these components is installed in most devices world-wide? (it's impossible to actually *know*)
| libcurl: | 719 |
| sqlite: | 381 |
| libz: | 443 |
| OpenSSL: | 488 |
Closed
@bagder I guess libcurl and sqlite are scoring very hig.
Given whos writing, probably libcurl, I guess ?
@bagder I guess it is not zlib/sql? You could technically be okay without those. SSL/libcurl not so much if you want to communicate (securely) with the internet.
@bagder that one was tricky, and the question is open for several interpretations. but with 100% sqlite, i guess my guess is correct
I deliberately chose to use the exact same wording for the question as when I ran the same poll back in late 2021.
Does Curl not use OpenSSL for https then?
@svavar often yes, but it also support ten other TLS libraries. Like curl.exe on all windows 10 and 11 for example, they don't use OpenSSL. Then there's of course the question if the openssl forks count as openssl or not...
@bagder Hmm, that's a tricky one, if we just count the number of instances I think sqlite is higher than libcurl or OpenSSL because those would likely depend on dynamically linked libraries installed once per machine, but sqlite is often embedded and thus installed multiple times... but you're asking specifically about which one is installed in most devices, not most installed overall.
@bagder if it's number of installs it's probably sqlite since we all have that installed multiple times per device but if each device is only worth one point I assume it's much closer.
My entirely unsubstantiated guess is that curl is much more common in IOT devices than sqlite and thus, wins here.
@bagder I voted sqlite, because there are more embedded devices that need a database than those that need networking. And out of those, not all use libcurl for networking.
The other two options I'm not sure. Every general-purpose device definitely ships with (likely several copies of) both.
@bagder I voted for OpenSSL. When BlackBerry 10 devices shipped they had three different versions of OpenSSL in them (all with different vulnerabilities of course).
@bagder I would've given the point to libcurl but it feels like every Android app uses sqlite. Also would libcurl even work without openssl? 🤷
@bagder i wonder why so many people bet on sqlite. IMHO it's the newest competitor in this game, and the are so many alternatives to it.
My bets are on zlib. I don't know any device that don't use an compressed kernel, compressed partitions, or libs.
@bagder Isn't libz basically a kernel dependency now days? So if you run linux you'd have it installed... unless you chose not to compress your kernel.
libcurl as a second if you want to talk over a network.
Then sql, cause everything needs a database now days.
Lastly OpenSSL because "security is out of scope for our LAN only thingamajig that's definitely never going to be connected to the Internet".
But I'm basing this on a guess that Linux use trumps network use, which might be naïve.
@bagder Can't be SQLite because my car's stupid infotainment unit keeps forgetting that I set it to "Shuffle"
@bagder Having seen Android App sources that contained multiple copies (and versions, of course) of some of these components in one single app, I guess the answer is all of them.
libcurl often pulls in libz and openssl, so its installs *may* inflate the other two. So I think it's *less* likely (but not impossible) to be #1.
libsqlite I don't have a vibe on. I know it's popular for embedded... 🤷♀️
OpenSSL has multiple alternatives on POSIX and Win/Mac have OS level crypto APIs. Embedded uses are probably better served by more... minimized, task-specific libraries.
zlib is the go-to IF interop (esp for legacy) is needed, but better algos exist for greenfield.
@bagder my guess is there are billions of mobile devices and they all need to store data, but not all of them need to curl something, so probably sqlite is the highest
I voted libz due to the dependencies in vcpkg:
curl depends on zlib . It also depends on openssl for most configurations that people actually use
sqlite3 depends on zlib as an option
I *thought* OpenSSL depended on zlib as well when I answered but it looks like I was mistaken
@bagder
I voted for OpenSSL because I assume that any serious use of libcurl also requires OpenSSL. This means that libcurl would always be used in combination with OpenSSL. Conversely, there are scenarios where OpenSSL is used without libcurl. This would mean OpenSSL is potentially used more frequently on devices worldwide.
However, there are also systems such as Microsoft Windows where OpenSSL is not present.
SQLite is very commonly embedded in systems, and libz is very often used as a dependency. It is really hard to say.
@bagder I answered libz because I used to work on non connected devices with need to compress resources in firmware, or with a PNG logo...
@bagder Que the meme where Java thinks it is the one at the top of the world, and then there are these libraries standing behind him, much more buffed.
If I understand correctly, they're using this module in particular:
https://git.madhouse-project.org/iocaine/nam-shub-of-enki
I also found a 'tutorial' for iocaine, which you might find useful:
https://winnerwind.in/tutorials/self-host/iocaine
The current state of password managers:
Bitwarden [1]
Vaultwarden [2]
KeePassXC [3]
1Password [4]
Proton [5]
[1] https://www.fastcompany.com/91542655/bitwarden-scrubs-always-free-and-inclusion-values-from-its-website-as-longtime-execs-step-down
[2] "A user you’ve blocked has previously contributed to this repository" (Claude)
[3] https://keepassxc.org/blog/2025-11-09-about-keepassxcs-code-quality-control/
[4] https://1password.social/@1password/116569353856438624
[5] Generally speaking about the company and also: "a user you’ve blocked has previously contributed to this repository" (Claude) in Proton Pass Github.
should i just stop caring about foss projects that use anubis?
i’ll probably migrate my projects to some self-hosted git forge that lets you browse the code without javascript on
@mkljczk Just to let you know that https://gitroot.dev has no JavaScript.
@manland i’m not anti-js, i’m developing javascript-heavy webapps, i just don’t want hard dependency on js in places where it shouldn’t be required and this is exactly what anubis does
@mkljczk seems like there's a lot of "use the slop program for bot protection to protect your slop from slop bots" happening
@Ember and i use noscript to protect myself from slop, including pre-llm era human slop (which is ironically the reason average internet users will continue to use llm chatbots as long as they don’t push advertisements in an overly obvious way)
That being said, I've been looking into alternatives, like iocaine, so I'm curious about your experience.
@me i never had a reason to run it but i don’t run public git forges anymore
Ciekawa interpelacja w Sejmie RP, a wśród pytań:
"6. Czy Ministerstwo Cyfryzacji rozważa utworzenie rządowej instancji platformy #Mastodon, wzorem rozwiązań wdrożonych przez Komisję Europejską, Niemcy i Holandię, jako oficjalnego kanału komunikacji administracji państwowej, zmniejszającego uzależnienie od komercyjnych platform społecznościowych? Jeśli tak, to kiedy? Jeśli nie, to z jakich powodów?"
Ja myślę, że w końcu powstanie, a Wy?
Co nie znaczy, że politycy odejdą z X.
Całość:
https://sejm.gov.pl/Sejm10.nsf/InterpelacjaTresc.xsp?key=DTEBLS
I keep seeing a lot of advice for writers saying "don't use this, don't use that" because it will show up as AI written.
I hate it. It makes me worried. A lot of the "AI" traits in writing overlap with storytelling language. Spoken language.
When I write, as a storyteller, I use em dash when I would take a breath or a pause on stage, instead of finishing a thought. I use triple adjectives because in spoken word they add to the rhythm.
1/2
I use "not only this - but also" for dramatic effect when I want to grab the audience's attention. The pause and the anticipation refocuses listeners.
I could go on, but you get the idea.
When I write, I like to bring the stories alive as if I'm telling them. A lot of storytellers do this in their books. And now people are like "don't do this if you don't want to be flagged as AI"???
We were here first, dammit.
@TarkabarkaHolgy We should stop using natural language because handling natural language is exactly the thing LLMs were invented for.
@navi I don't think it was some engineered plan to centralize development, I'm pretty sure it was just an unfortunate sequence of events stemming from the lack of a stable abi and wanting to follow javascript's model. There are alternative crate hubs, but its like saying there are alternative bluesky backends. In the end, they all loop back to the central hub, which is unsustainable.
@navi Oh, I absolutely agree with that. crates dot io has caused a lot of supply-side issues with distributing Rust programs and is a huge security vector because nothing is audited. Personally, I much prefer Hare's distribution philosophy, as, even though it is a statically linked language, it does not have a central distribution hub and instead expects source-code of libraries to be packaged, which the compiler picks up at build-time. This is sort-of how Debian handles Rust as well.
@navi I can see Rust reaching that point eventually, because all the pieces are there (see Debian's packaging of Rust programs), but this was also partially the fault of traditional UNIX-like systems. There wasn't much infrastructure or understanding in place to be able to handle packaging a large language enforcing static linkage, so the easiest path of using crates dot io for dependencies was the most common solution. I don't see much changing in the sort-term though.
RE: https://mstdn.social/@rysiek/116573743512232707
Temat weryfikacji wieku online jest ważny, wpis na blogasku awansował więc na artykuł w OKO.press:
https://oko.press/weryfikacja-wieku-dzieci-w-social-mediach-nie-ma-bezpiecznego-sposobu
Weryfikacja wieku dotyczyć będzie każdego, nie tylko dzieci.
Jej implementacja wymagać najpewniej będzie smartfona z systemem jednego z duopolistów.
Katalog stron nią objętych będzie z czasem rozszerzany.
Nawet najlepiej zaimplementowana dalej będzie problemem dla prywatności.
I będzie wymagać filtrowania Internetu, kontroli aplikacji, i zamkniętych systemów operacyjnych.
Nie ma weryfikacji wieku bez filtrowania Internetu:
https://rys.io/pl/183.htmlWeryfikacja wieku online jest problemem wysoce nietrywialnym. Jest też po prostu złym pomysłem.
Nawet z najlepszym, najbardziej chroniącym prywatność systemem, weryfikacja wieku nadal będzie pomagała identyfikować nas w sieci.
I wreszcie: nie da się wdrożyć skutecznej weryfikacji wieku bez filtrowania Internetu. A ten temat w Polsce wałkowaliśmy już tak wiele razy, że naprawdę powinniśmy wiedzieć lepiej.
Jeśli problemem, który chcemy rozwiązać, są krzywdy wyrządzanie dzieciom przez VLOPy, to skupmy się na tych krzywdach i odpowiedzialności ich operatorów.
To, że VLOPy są toksyczne, nie jest przecież przypadkowe.
Jest świadomą, celową decyzją ich operatorów. Taka decyzja nie jest konieczna dla świadczenia podobnych usług.
Usługi Mety toksyczne są, dlatego, że dzięki temu Mark Zuckerberg więcej zarabia. A my znów debatujemy o filtrowaniu internetu, zamiast o rozwiązaniu problemu u jego źródła.
@rysiek to powinno skończyć się sprawami przed TSUE i w polskich sądach. Nie wyobrażam sobie żeby w celu dostępu do ulubionych stron trzeba było korzystać z telefonu z systemem Google lub Apple.
@rysiek Dzięki za wysiłek włożony w informowanie o tym problemie rysiek!
Temat rzeczywiście ważny, i mam wrażenie że za każdym razem jak coś na ten temat czytam, dowiaduję się o kolejnym argumencie, dlaczego weryfikacja wieku w Internecie jest złym pomysłem.
Powierzchownie temat tego kalibru wydaje się oczywisty, platformy/treści w internecie są złe, dzieci trzeba chronić, wydaje się oczywiste że rozwiązaniem jest wprowadzenie weryfikacji aby „odfiltrować” dzieci z sieci. Komuś, kto nie ma wiedzy technicznej bądź czasu aby zapoznać się ze szczegółami łatwo będzie machnąć łapą i zgodzić się z takim rozwiązaniem.
Dlatego myślę że tak ważne jest informowanie społeczeństwa o konsekwencjach proponowanych legislacji.
Dzisiaj argumentów przeciwko weryfikacji wieku jest od groma, zarówno z perspektywy technologicznej, społecznościowej, etycznej, logistycznej, statystycznej. Jak do kogoś nie trafiają argumenty jednej kategorii, powinny trafić z kategorii dla tego stworzenia bliższej.
Mam nadzieję że głosy osób negatywnych w stosunku do proponowanych zmian będą na tyle głośne aby przedrzeć się przez mainstreamowy dyskurs i przynajmniej dać niektórym możliwość refleksji.
@rysiek Przy okazji, na prawdę zastanawiam się czy dzisiaj w przypadku wprowadzenia szkodliwych legislacji, możliwa byłaby odpowiedź społeczeństwa jak ta przy wprowadzaniu ACTA. Mam szczerą nadzieję że tak.
@rysiek tyle, że to nie jest do końca prawda. W końcu czy ktoś musi zawsze korzystać z mediów społecznościowych na smartfonie? Nie, a na kompie nie ma żadnych ograniczeń odnośnie systemu.
@T_Wezmyr przeczytałeś tekst? Bo jeśli byś przeczytał, to na przykład dowiedziałbyś się, że aplikacja do weryfikacji wieku jest tylko mobilna.
Więc dla 99.99% ludzi (poza osobami, które sobie odpalą w emulatorze) smartfon byłby obowiązkowy.
Polecam lekturę.
@rysiek taaaak, dlatego właśnie pisze, że aplikacja jest bez sensu. A m Obywatel ma sens, bo raz, że już jest, dwa że działa wszędzie.
@rysiek W social mediach? Właśnie odkryłem potwierdzenie tożsamości w sklepie internetowym, i nie, nie sexshop
I've done the right thing and it's going to cause pain.
#Gentoo Distribution Kernel configs are now hosted entirely on git.gentoo.org rather than GItHub. If you don't use Gentoo mirrors, you may be hitting 502s thanks to our LLM overlords now. If you use Gentoo mirrors, you may be hitting 404s if they hit 502s while trying to fetch from our Infra 🤷.
@navi what i didnt know this sobs qwq ,😭😭😭😭
the last weeks we saw more and more security issues coming up. Let's talk!
Sorry, a pretty long blog post about this...
https://gyptazy.com/blog/coding-after-ai-are-humans-still-good-enough/
#ai #aicoding #coding #opensource #foss #security #infosec #vulns #developer #devops #engineer #ops #fedi #philosophy
@gyptazy
Except that we write software primarily to solve human problems: to build a building, manage money, buy or sell goods. What does AI know about real human problems, not in the past, but now and in the future? How can we guarantee that the code we write follows our needs and respects the law?
And if AI messes up, who is responsible? Because when global, massive, automated systems mess up, they mess up badly.
This is your official notice that we'll be making a major security release for Sharkey on the 20th of May, 2026, between
2026-05-20T13:00:00.000Z
and
2026-05-20T15:00:00.000Z
. Prepare to update immediately upon release. If you are unable to update your instance at the given time, take it offline before the release.
I see 1Password is stepping on the rake again.
The brain geniuses securing your passwords just squirted out this statement: "We applied agentic tooling to a multi-million-line Go monolith."
Ok first, how in the everloving fuck did you allow a *password manager* to reach *multiple millions of lines*??
Welp, I warned about 1P's self-immolation in 2023 and posted a migration guide. You're welcome.
DO NOT reply telling me what manager you're using instead. I don't care.
@jwz I currently self-host Vaultwarden, a Bitwarden-compatible server written in Rust.
@ewhac I didn't yet find where this comes from, but:
https://metroholografix.ca/@sb/116573522625745018
@jwz Very happy with VaultWarden
@chris I did not yet find where this comes from, but:
https://metroholografix.ca/@sb/116573522625745018
@viq That sounds like a reasonable stance.
@chris that depends whether someone's dislike of LLMs is based on their results, or on how they came into being and what's necessary for their existence and operation 🤷
@viq I don't mind using LLMs to contribute to OpenSource from a "sourcing" of training data standpoint that much. I agree though that the resource usage is currently just unsustainable. But so is it for the LLM companies. At a certain point the costs will catch up and then the price for tokens/subscriptions will go through the roof. That is the point when unsolicited LLM PRs will probably drop in number. I believe that time is near.
@chris and I've seen who is creating those tools, how and why, and want to have nothing to do with all of that.
No, that's not entirely accurate 🤔 I think there are circumstances in which spending some time with the people who are driving those processes could actually be productive, to let them know how it's affecting society and how we feel about it.