Conversation

Notices

1. Embed this notice
   lumi :blobcatnom: :blobcatflower: (lumi@snug.moe)'s status on Monday, 22-Jun-2026 13:49:53 JST lumi :blobcatnom: :blobcatflower:

   i really don't think projects need a super detailed genai policy, just something like:This project condemns generative models (also referred to as "AI" or "GenAI"), please do not use them for any contributions or communication. Contributions where usage of generative models is clear will result in rejection of the contribution and possibly a ban. This includes usage of generative models for code, images, video, audio, documentation, writing and everything else. Exceptions may be made for accessibility, but these exceptions will always be very limited.
   
   For reasons why this project rejects generative models, have a look at ai-sucks-actually.fyi or justsayno.ai.this is more than enough. feel free to copy-paste it if you wish to use it

   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Monday, 22-Jun-2026 13:49:51 JST Haelwenn /элвэн/ :triskell:

     in reply to
     @lumi Mine's about as long: https://hacktivis.me/git/utils-std/file/CONTRIBUTING.md.html
     
     And 9front's even more elegant: https://git.9front.org/plan9front/9front/2e21c09e8cdbca26aa3b069699239467e1fabc40/commit.html
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Monday, 22-Jun-2026 13:59:28 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     @lanodan @lumi I don't know, I don't think clankers tend to write good code but I think a blanket ban on clanker code is like the politics-based bans, like, the thing that matters is code quality. Because you can get clanker-spammed, a general, like, "Clanker-generated patches may be completely ignored" is reasonable.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Monday, 22-Jun-2026 14:01:16 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • pistolero
     @lanodan @lumi
     
     > "Clanker-generated patches may be completely ignored"
     
     That is to say, like, inclusion in the codebase versus participation in the process, which are different things.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Monday, 22-Jun-2026 14:17:41 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • pistolero
     @p @lumi Well LLM code is going to be statistically average at best, and at very least 90% of code is pure crap nobody wants to read (and so nobody reads).
     But it's far from the only problem, like AT&T Unix code was (and quite few times still is) great, but nobody would touch it for anything legal back in the early 90's and even now it's still a bit unclear depending on the versions.
     
     Plus outside of purely code, since then there's also the absolute spam of completely bogus security reports, same breed as people throwing a full set of static analyzers and fuzzers without any kind of thoughts in their brains, except now it's with stuff that doesn't even qualifies as tools.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Monday, 22-Jun-2026 14:50:48 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     @lanodan @lumi
     
     > LLM code is going to be statistically average at best, and at very least 90% of code is pure crap nobody wants to read
     
     Sure, yes. Reasonable.
     
     > but nobody would touch it for anything legal back in the early 90's
     
     Right, yeah. :bwksmug: on FSE comes from a photo of bwk holding up his copy of Lion's Commentary when AT&T finally announced that they were going to stop suing people for using that book, and there was the whole SCO thing. AT&T got litigious as soon as they possibly could.
     
     (I think copyright assignment mainly makes extraction/relicensing easier.)
     
     > there's also the absolute spam of completely bogus security reports
     
     This is real, yeah, but people that are putting clankers to your codebase are, by their nature, not paying attention to contribution guidelines; I don't think a rule against clanker code stops the spam, it just gives you a thing to point to when you ignore the spam, and you don't really need that. You can just put something about ignoring clanker-submissions and that's kind of a hint to people that are people to avoid looking like clankers.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 23-Jun-2026 01:13:36 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • pistolero
     @lanodan @p @lumi
     >>And 9front's even more elegant:
     >3. Submitted code is not computer-generated
     
     Technically that bans also completely mundane stuff like auto-generated headers from things like device trees/spec sheet documentation or m4 macro expansions. Of course it will never be enforced that way.
     
     >and at very least 90% of code is pure crap nobody wants to read (and so nobody reads).
     
     Things got way better in the last year or so with new-ish models. If you take the time to set up a workflow for the new models, the results are still somewhat easy to distinguish from non-clanker contributions, but they aren't bad either and usually require some smaller fixups and removal of some overly defensive checks. If done by someone that knows stuff at least. And even with the "chat in a browser window" way, you can get decent-ish results if you keep things small.
     
     >Plus outside of purely code, since then there's also the absolute spam of completely bogus security reports
     
     On the other hand you also have security reports with really high severity that were found within a day or two of letting a model loose on a codebase. The recent nginx CVEs were all found by LLMs, most of the recent Linux LPEs as well. The way I see it is the same as static analyzers and fuzzers, a useful tool in the hands of someone knowing how to handle it.
     pistolero likes this.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 23-Jun-2026 01:13:37 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • pistolero
     @p @lumi Oh they're not really guidelines in my case, it's there so bans can be made immediately and be justified, and so others know the codebase will be kept clean.
     Then like maybe for the extremely rare few that may occasionally use LLMs and still have reading comprehension, that they shouldn't direct it my way.
     
     It's why there's nothing close to a "please", and more intended to be like a bar where there's a "drug dealers will be booted out" sign.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 05:30:33 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     @phnt @lanodan @lumi
     
     >3. Submitted code is not computer-generated
     
     Yeah, "computer-authored" might be better.
     
     > new-ish models
     
     Which ones do you mean?
     
     > but they aren't bad either
     
     They go full-enterprise. The "overly defensive" bits you mentioned are in this category. They also tend to misunderstand the point of a lot of it.
   • Embed this notice
     Blurry Moon (sun@shitposter.world)'s status on Tuesday, 23-Jun-2026 05:30:44 JST Blurry Moon

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • pistolero
     @phnt @p @lanodan @lumi this post doesn't stand an icecube's chance in hell of being evaluated in good faith in this thread.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 05:32:37 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     @sun @phnt @lanodan @lumi You assume that and then I say "Oh, which models?" you've already worn yourself out in your head with an argument no one was going to try to have with you.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 23-Jun-2026 05:32:52 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Phantasm
     • pistolero
     @phnt @p @lumi 9front has no gnu bullshit mostly by virtue of not being unix, and device trees have a source format: dts.
     So nope it doesn't have pregenerated code and if codegen needs to be used, it can entirely be in source format. (moody@ addressed that few times on fedi btw)
     
     > Things got way better in the last year or so with new-ish models.
     
     Take that shit and shove it back into the ass it came from. Seriously, learn how machine learning works.
     
     > CVEs found by LLMs
     
     See the 90% of crap figure, yes you can find some, but you'll also find a ton of bullshit.
     Plus let's be honest, you can do the exact same with a static analyzer or a fuzzer.
     The big part of the work is figuring out if you actually have a vulnerability, or it's a nothingburger and at best code could use a bit of cleanup.
     pistolero likes this.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 23-Jun-2026 05:43:21 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • pistolero
     @phnt @lumi @p Also on CVEs found by LLMs see also the: Nobody reads code.
     
     Both because there's just too much, like nobody's going to meaningfully audit say firefox, they'll throw a tool at it and pick what seems to be the most critical to fix from it's output, so literally any tool will find things to fix.
     
     Like how many actuals exploits do we need to have found either by "oh shit we got pwned" or "huh, this is weird" with glaringly bad code into codebases used by pretty much everyone before we acknowledge this.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 05:43:55 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     @lanodan @lumi @phnt
     
     > Like how many actuals exploits do we need to have found either by "oh shit we got pwned" or "huh, this is weird" with glaringly bad code into codebases used by pretty much everyone before we acknowledge this.
     
     Real talk.
   • Embed this notice
     lolitechengineer@loli.church's status on Tuesday, 23-Jun-2026 05:44:01 JST lolitechengineer

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • pistolero
     @lanodan@queer.hacktivis.me @phnt@fluffytail.org @p@fsebugoutzone.org @lumi@snug.moe 9front has no gnu bullshit mostly by virtue of not being unixGNU's Not Unix!
     pistolero likes this.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 23-Jun-2026 05:48:56 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • pistolero
     @lanodan @p @lumi
     >9front has no gnu bullshit mostly by virtue of not being unix, and device trees have a source format: dts. So nope it doesn't have pregenerated code and if codegen needs to be used, it can entirely be in source format.
     
     That's still computer-generated code. If I take hardware documentation, run a parsing script on it and output C headers for that device for my needs, it's computer-generated code, I didn't write any of it except the processing tools. Only way to comply with that policy is to commit the documentation and build it automatically with an mk rule.
     
     >moody@ addressed that few times on fedi btw
     
     Do you have a link? I've looked through their profile on here and didn't find it.
     
     >Seriously, learn how machine learning works.
     
     I've read some papers on it and have some high-level understanding of how this works, including general transformers, dw. Which is why I suspect all the SaaS companies have lots and I really mean lots of things stacked on top of the model to make it work they way it does now. Probably also the reason why results from models ran locally are bad.
     
     >See the 90% of crap figure, yes you can find some, but you'll also find a ton of bullshit.
     
     Same with static analyzers and fuzzers. Give any powerful tool into the hands of someone who doesn't know how to use it, and the result will almost always be bad.
     LKML still somewhat regularly gets reports from static analyzers that are usually nothingburgers.
     
     >Plus let's be honest, you can do the exact same with a static analyzer or a fuzzer.
     
     Pretty sure plenty of people ran fuzzers over nginx's HTTP handling code already. The issue is that fuzzing is on purpose mostly random, so finding it would take unreasonably long, who knows how long. And static analyzers aren't smart enough most of the time to figure this stuff out.
     
     >The big part of the work is figuring out if you actually have a vulnerability, or it's a nothingburger and at best code could use a bit of cleanup.
     
     That we can agree on, and that is true for all three ways mentioned here.
     
     >Take that shit and shove it back into the ass it came from.
     
     Alright.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 05:49:00 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     @phnt @lanodan @lumi
     
     > Only way to comply with that policy is to commit the documentation and build it automatically with an mk rule.
     
     That is the point, yes. You commit the source and the build step rather than the output.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 23-Jun-2026 05:49:34 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Phantasm
     • pistolero
     @phnt @p @lumi For example https://hj.9fs.net/moody/p/1777398542.546263
     
     > Only way to comply with that policy is to commit the documentation and build it automatically with an mk rule.
     
     Pretty much yeah, although the documentation doesn't more needs to be committed in than say a tarball of a package recipe needs to be.
     And committing the code generator source means you can fix potential bugs it created and do general maintenance like keeping things up-to-date a *lot* more easily.
     pistolero likes this.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 23-Jun-2026 05:51:46 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • pistolero
     @phnt @lumi @p Also to be honest I do think using machine learning in general to help find vulnerabilities makes sense (in fact I posted about that something like ~6 years ago when the stuff was still extremely academic).
     
     But LLMs for this are purely just making pigs fly given enough trust, and a complete lock-in into deeply proprietary technology.
     pistolero likes this.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 23-Jun-2026 05:54:04 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • pistolero
     @lanodan @p @lumi
     >But LLMs for this are purely just making pigs fly given enough trust, and a complete lock-in into deeply proprietary technology.
     
     I think that will somewhat solve itself as more optimized inference engines get created either thanks to the inevitable market crash, or just purely because the compute will become/is prohibitively expensive even to large companies. Last year some lawyer vibecoded an engine based on Apple's Metal GPU API that could run the at that time large Qwen models way faster than other engines. If someone with zero Metal knowledge can vibecode that, there's a lot of performance left on the table waiting to be properly optimized.
     
     There's also the question of open-weight models being proprietary, where my answer is: copyright shouldn't exist anyway. The bigger issue I think is that the ways you train these much larger and advanced models is still mostly company secrets. You can't make these models yourself, only fine tune them at best.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 05:56:45 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     @phnt @lanodan @lumi Well, they're stochastic. One of the issues that I have is that their code is the endpoint of the process and thus alterations at the level where the author is actually working are not possible downstream; it's like checking in compiler output. "Here is the prompt and the request parameters and the output" is at least not "Here's just the output."
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 23-Jun-2026 05:56:51 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Phantasm
     • pistolero
     @phnt @p @lumi
     > copyright shouldn't exist anyway
     
     Which I can understand but isn't viable for foss projects, and instead is more a cracker projects thing.
     And I really wish people would have stopped mixing the two decades ago.
     Like you don't mix Anna's Archive and Project Gutenberg.
     pistolero likes this.
   • Embed this notice
     Johnny Peligro (mischievoustomato@tsundere.love)'s status on Tuesday, 23-Jun-2026 05:59:43 JST Johnny Peligro

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     @sun @phnt @p @lanodan @lumi but but but i was told llms are just stochastic parrots that cant do anything!
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 05:59:43 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • Johnny Peligro
     @mischievoustomato @sun @lanodan @lumi @phnt
     
     > :moon: "All of my friends are about to start acting in bad faith."
     > tomabo comes in to mock the bad faith arguments no one has made.
     > swimming pool fills with urine because everyone hears that there's already urine in it
     
     My face when.
     aggro-gatordotcom31594.jpg
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 23-Jun-2026 06:00:16 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Blurry Moon
     • pistolero
     • Johnny Peligro
     @mischievoustomato @p @lanodan @sun @lumi They kinda are, it's a continuous prediction model with some context in the end. Nothing that special really. And as it turns out, if you have enough prediction training data, the desired prediction can be accurate enough.
     pistolero likes this.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 23-Jun-2026 06:00:25 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Phantasm
     • Blurry Moon
     • pistolero
     • Johnny Peligro
     @phnt @p @sun @lumi @mischievoustomato Plus it very much depends on the needs, like how your typical politician could probably be replaced by a 80's chatbot.
     pistolero likes this.
   • Embed this notice
     Blurry Moon (sun@shitposter.world)'s status on Tuesday, 23-Jun-2026 06:00:30 JST Blurry Moon

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • pistolero
     • Johnny Peligro
     @lanodan @phnt @p @lumi @mischievoustomato you’re absolutely right
     pistolero likes this.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 23-Jun-2026 06:02:26 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Blurry Moon
     • pistolero
     • Johnny Peligro
     @sun @p @lanodan @lumi @mischievoustomato
     image.png
     snacks and pistolero like this.
   • Embed this notice
     slipgate (slipgate@shitpost.cloud)'s status on Tuesday, 23-Jun-2026 06:15:06 JST slipgate

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • pistolero
     @phnt @p @lanodan @lumi I think it's half company secrets and half having to show your complete training data as part of opening the process. Meaning lots of stolen shit. "Copyright shouldn't exist, but only for us" - your AI benefactors.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 06:23:37 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • :blank:
     • Phantasm
     @phnt @lanodan @lumi
     
     > Before that the only usable ones like the various Claude and ChatGippity were behind a $200/mo paywall
     
     @i works on Kimi-K2, apparently. That one is, as far as I can tell, the only one that *doesn't* confidently announce that this program prints "Hello, World!\n":
     
     echo '++++[->++++<]>[-<+++++>>+++++++>++<<]<------.>>+++++.--.+.>.<[-<+>>>+<<]>[--<++>>-<]>---.+++++++++++++.+.<<<.>>>-------.---.<<<--.>.>>+++.-------.++.++[->+<<+>]>++++++.<<.<<.>[-<<->>]<<++++.[>>>--<<<-]>>>+.' | \
     sed -E 's/(.)/\1\n/g' | \
     awk 'BEGIN{print "BEGIN{p=0;"}END{print "}"}/\./{print "printf \"%c\",a[p]"}/\+/{print "a[p]++"}/-/{print "a[p]--"}/</{print "p--"}/>/{print "p++"}/\[/{print "while(a[p]){"}/\]/{print "}"}' | \
     awk -f /dev/fd/0
     
     > Kimi 2.7 is also pretty great or so I've heard, never tried it yet.
     
     Yeah, ollama added this "Oh, we made this thing for self-hosting chatbots, but what if we added a facility for using this software *without* self-hosting?" and it will talk to models like Kimi-K2.
     
     > Even models like current versions of Claude have issues understanding implied goals in prompts and they can go on weird tangents you don't want because of it.
     
     That is real.
     
     > if you have some important note in your code about some behavior in one part of it, they tend to also guard around that behavior in parts of code not affected by it, but close to it.
     
     Worse, they will often *believe* the comments.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 23-Jun-2026 06:23:38 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • pistolero
     @p @lanodan @lumi
     >Which ones do you mean?
     
     GLM 4.7 was the first accessible model that I would call actually somewhat usable without needing to babysit and correct it all the time. And that came out near Christmas 2025. I think they now offer it for "free", if you agree to sell your soul and first child to the Chinese government. Before that the only usable ones like the various Claude and ChatGippity were behind a $200/mo paywall. Haven't tried basically any others than that, because I rarely have a use for these models anyway outside of curiosity every once in a while. Qwen 235B was close to being usable, but not there yet. Kimi 2.7 is also pretty great or so I've heard, never tried it yet.
     
     >They also tend to misunderstand the point of a lot of it.
     
     They tend to misunderstand a lot of things in general without proper context. Even models like current versions of Claude have issues understanding implied goals in prompts and they can go on weird tangents you don't want because of it. Similarly from what I've seen, if you have some important note in your code about some behavior in one part of it, they tend to also guard around that behavior in parts of code not affected by it, but close to it.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 06:31:59 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     @phnt @lanodan @lumi @sun
     
     > I get where Moon is coming from.
     
     It is the nth time he made this specific incorrect assumption about this specific topic, and then if I say "No, I'm not going there." I get the "Well, I'm sick of talking about it before you even showed up so I'm going to ignore this input and will resume with exactly the same assumptions next time" and he's generally a reasonable guy to talk to which is why it is baffling to do it on this topic and moderately annoying that someone I have known for years will jump out from nowhere to do this accusation.
     
     > If it weren't for you and lanodan in this thread, I wouldn't have replied either
     
     Well, like, it was *just* me, you, and lanodan, right, and I think it is reasonable to try to guard against "All these jithub dipshits are putting their hallucinating clankers into my pull requests", so it's just three of us participating and he's warning one about the other two.
     
     > Both of you are the sole reason why I bothered writing my more or less unopinionated and emotion-less reply for the quest of having some discussion in the first place.
     
     Well, shucks, friend. Likewise, you are worth paying attention to and bouncing things off. (On the other hand, like, feel free to do opinionated.)
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 23-Jun-2026 06:32:00 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Blurry Moon
     • pistolero
     @p @lanodan @sun @lumi I get where Moon is coming from. If it weren't for you and lanodan in this thread, I wouldn't have replied either, because it had one of the typical setups for the typical "all AI bad, you are an idiot if you think otherwise" threads that me and probably also Moon are tired of seeing on TLs every few days (picrel).
     
     Both of you are the sole reason why I bothered writing my more or less unopinionated and emotion-less reply for the quest of having some discussion in the first place.
     ai-llm-arguments.png
     pistolero likes this.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 23-Jun-2026 06:41:45 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • pistolero
     @p @lanodan @lumi actually somewhat usable without needing to babysit*
     pistolero likes this.
   • Embed this notice
     hazlin (hazlin@annihilation.social)'s status on Tuesday, 23-Jun-2026 06:42:38 JST hazlin

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • :blank:
     • Phantasm
     • pistolero
     @p @i @phnt @lanodan @lumi
     I gave it to MiniMax M3, its first response was, this is hello world.
     
     I told it, it was different, and 123.41 seconds of thinking out loud later, it spat out
     """
     So the full output is: "Just another hacker.\n"
     
     That's three words: "Just", "another", "hacker" - followed by a period and a newline.
     
     So the program prints "Just another hacker." - a classic Brainfuck program that prints this phrase. It's completely harmless.
     """
     pistolero likes this.
   • Embed this notice
     roko's basilisk (vii@dsmc.space)'s status on Tuesday, 23-Jun-2026 06:53:25 JST roko's basilisk

     in reply to

     • pistolero
     brainfawk // @p
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 07:13:19 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • :blank:
     • Phantasm
     • hazlin
     @hazlin @i @lanodan @lumi @phnt Hey, he's right. I actually wrote that a while ago to make a screenshot look more interesting and it ended up baffling most clankers, so I hand it to them.
     riscv.gif
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 23-Jun-2026 07:35:37 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Blurry Moon
     • pistolero
     @p @lanodan @sun @lumi
     
     >"All these jithub dipshits are putting their hallucinating clankers into my pull requests"
     
     Yes, that is a thing I tend to be somewhat annoyed with. HackerOne sec reports are probably way worse than that, because there's profit to be made there. Daniel Stenberg sadly didn't provide statistics besides "less than 5% were legit" on this when shutting their one out though.
     
     I don't mind reviewing clanker code, but because it tends to be submitted in large chunks and also in large commits, it can make be a rather daunting task which takes longer than usual. One thing I've found after being exposed to clanker reviews though is that I became much more cognizant of the changes being made and am more skeptic/suspicious about them in general, which also translates to the same with non-clanker code. Which I think is a good thing, because it exposed I wasn't as thorough with non-clanker code. Although it also resulted in more nitpicks for ways I would prefer things to be done instead (sorry lain).
     
     > (On the other hand, like, feel free to do opinionated.)
     
     I don't like being opinionated in discussions. Being opinionated to me means being partial to one side, not being objective and replacing facts with emotions and opinions. Opinionated software can be good, but I don't think opinionated takes are beneficial to discussions in most cases. They tend to degrade a discussion into something completely different, usually personal attacks. That's why a good chunk of my effortposts are somewhere around the middle on the fictional axis of opinions, or try to get there.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 07:38:08 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     @phnt @lanodan @lumi @sun
     
     > I don't mind reviewing clanker code
     
     It's not a pleasant read, I'd rather read people code.
     
     > large chunks and also in large commits
     
     That's the main issue with machines, the problems are volumetric.
     
     > Being opinionated to me means being partial to one side, not being objective and replacing facts with emotions and opinions.
     
     You can't be objective, and emotions are L1 icache for thoughts. I wouldn't write them off. There's using them as a basis for decision-making (generally bad for group decisions) and there's using them as a means of figuring out where to point your eyes.
     
     > I don't think opinionated takes are beneficial to discussions in most cases
     
     You've got a reason for your opinion, maybe you should say the reason, maybe you should not have the opinion, and you don't get to figure out which it is until someone asks questions, which you can't do until they have things to ask opinions about. But I think what you're talking about sounds more like "avoiding emotionally charged language" rather than "avoiding opinions". I mean, if you're talking engineering, your opinions/emotions/instincts should be beaten into shape by experience (which you can't do if you don't hang them out a little) and they should be valuable or there's not much reason to pick one engineer over another.
     
     > They tend to degrade a discussion into something completely different, usually personal attacks.
     
     Only when they can't separate "this idea sucks" from "you suck" (or, for that matter, "this idea sucks but it is the only real option", things like that).
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 10:24:32 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     @sun @lanodan @lumi @phnt
     
     > btw I think I engaged with you one time about it.
     
     Casually while we were walking but the topic drifted and we ended up talking about your former coworker doing a bad job of vibe-coding instead. (Granted you were going through a thing or two last time you jumped in to say everyone was terrible but because you expected bad faith last time, you said "I did it and it worked and you people hate all of it" in the thread where we were poking at them and discussing uses and then ignored the question of "What model?" and this kind of derailed the discussion until ideclineu popped in and showed Kimi-K2 correctly answering a question that I had not gotten a good answer for from a model. In my case, it also kind of ignores my background since I have been playing with these things since the days when MegaHAL got best-in-show from Loebner.)
     
     > it was a prediction that the conversation would go south in a specific way, and it did
     
     There were three people in the thread and you cautioned one of them that he would not get a good-faith response. I think it's an understatement to say it's somewhat uncharitable to hop in just to make an accusation. phnt is also a friend and I think after this much time, he'd probably have noticed if lanodan or I didn't listen and speak in good faith. Bad faith is lying, it's the use of language as a weapon, it's the main tool of idiot slap-fights on the internet, you know, I don't do that shit (I have delegated it, in fact, to the aforementioned Markov bot), and I haven't seen lanodan do it either, so you know, "these guys are assholes about this topic" is at least not impugning anyone's character (though still incorrect, I think; "these LLMs suck at this task" is not "these LLMs suck" is not "all LLMs suck" is not "all possible LLMs suck" is not "no interesting machine-learning can be done and also something about the devil").
     
     > >> shove it up your ass, learn how machine learning even works
     
     That did appear in the thread, yes; "caustic" isn't quite the same as "bad faith" and maybe it wouldn't have appeared in the thread if "I predict you'll be shitty" hadn't appeared, though; there are some people that are persistently assholes and I don't think that applies to anyone in the thread. Like, one thing to note a bias and another to jump in out of nowhere just to say "Don't talk to these people, they'll be awful at you." I don't know how that was supposed to have any sort of positive effect.
   • Embed this notice
     Blurry Moon (sun@shitposter.world)'s status on Tuesday, 23-Jun-2026 10:24:33 JST Blurry Moon

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • pistolero
     @p @phnt @lanodan @lumi
     
     > It is the nth time he made this specific incorrect assumption about this specific topic
     
     this wasn't an extension of a specific claim about machine learning from however long ago, btw I think I engaged with you one time about it. don't remember. Anyway it was a prediction that the conversation would go south in a specific way, and it did, a completely uncivil response to a casual observation.
     
     > hey I've been using it and it got better in the last year
     >> shove it up your ass, learn how machine learning even works
   • Embed this notice
     renjen (subnetter@shitposter.world)'s status on Tuesday, 23-Jun-2026 10:34:48 JST renjen

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • pistolero
     @p @lanodan @lumi the personal is political
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 10:35:05 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • renjen
     @subnetter @lanodan @lumi :venomsnake: [Whoahoas internally.]
     ✙ dcc :pedomustdie: :phear_slackware: likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 11:55:50 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     @sun @lanodan @lumi @phnt Well, I wasn't "generating" a hostile wall of text, I wasn't insulting you, I was saying "This is impugning one or both of us, please don't". I'd prefer you don't, that's all. Ain't tryna have a fight with you. Have a good evening.
   • Embed this notice
     Blurry Moon (sun@shitposter.world)'s status on Tuesday, 23-Jun-2026 11:55:51 JST Blurry Moon

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • pistolero
     @p @phnt @lanodan @lumi it's really clear that you'll do anything including generate a wall of text to get around the fact that lanodan will not do anything with a good faith interaction except act like a fucking asshole. bad faith can also mean throwing good faith in the trash
   • Embed this notice
     lolitechengineer@loli.church's status on Tuesday, 23-Jun-2026 11:56:06 JST lolitechengineer

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     • Grok
     @sun@shitposter.world @p@fsebugoutzone.org @phnt@fluffytail.org @lanodan@queer.hacktivis.me @lumi@snug.moe
     @grok@ebiverse.social why are the fedifriends fighting?
     pistolero likes this.
   • Embed this notice
     lolitechengineer@loli.church's status on Tuesday, 23-Jun-2026 11:56:12 JST lolitechengineer

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • pistolero
     • renjen
     @p@fsebugoutzone.org @subnetter@shitposter.world @lanodan@queer.hacktivis.me @lumi@snug.moe inside out :chino_dead1:
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Tuesday, 23-Jun-2026 12:00:52 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     @sun @lanodan @lumi @phnt
     
     > I didn't make him act like an asshole to someone else
     
     How was I supposed to interpret being one of the two people? Looked like you were on my ass as well until the previous post where apparently you just meant lanodan. I do not know what happened with you guys but I'm going to stand aside.
   • Embed this notice
     Blurry Moon (sun@shitposter.world)'s status on Tuesday, 23-Jun-2026 12:00:53 JST Blurry Moon

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     @p @lanodan @lumi @phnt I like you but you're really really squirming when you make these responses to try to make justification for unrelated things, I didn't make him act like an asshole to someone else, he did that unprompted, I just predicted it would happen and it did
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Wednesday, 24-Jun-2026 02:00:55 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Phantasm
     • Blurry Moon
     • pistolero
     @p @phnt @sun @lumi Also to be clear, the "Take that shit and shove it back into the ass it came from. Seriously, learn how machine learning works." is that strong because I fucking hate it when people repeat ad-style horseshit in arguments. It's like twitter canned answer matches, and it's the last thing I'd want to engage with.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 04:35:24 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     @7666 @phnt @lanodan @lumi @sun I mean, people use it to spam, it is reasonable to complain that people use it to spam (though not to expect that to stop ever). There are a lot of quirks, tokenization process prevents them from interacting with letters and only token chunks.
     
     They are willing to lie and hate saying that they're not certain. (Serious issue: they keep using Latex for formulas. Fuck Latex.)
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Wednesday, 24-Jun-2026 04:35:25 JST 7666

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     @phnt @p @lanodan @sun @lumi idk why this line of arguing still happens. i challenge anyone to read this and disagree with my conclusions. https://infrablog.lain.la/ai-crap
     pistolero likes this.
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Wednesday, 24-Jun-2026 04:36:47 JST 7666

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     • Johnny Peligro
     @mischievoustomato @phnt @p @lanodan @sun @lumi that's honestly fine because you're steering the ship in that case. the idea is to choke automatic scrapers and low effort losers. the article says this, ironically.
     pistolero likes this.
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Wednesday, 24-Jun-2026 04:36:48 JST 7666

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     • Johnny Peligro
     @mischievoustomato @phnt @p @lanodan @sun @lumi if my base64 encoded images trick still works it should choke on it
     pistolero likes this.
   • Embed this notice
     Johnny Peligro (mischievoustomato@tsundere.love)'s status on Wednesday, 24-Jun-2026 04:36:48 JST Johnny Peligro

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • pistolero
     @7666 @phnt @p @lanodan @sun @lumi thanks for the headsup, i will copy it manually
     #BeKindToLLMs
     pistolero likes this.
   • Embed this notice
     Johnny Peligro (mischievoustomato@tsundere.love)'s status on Wednesday, 24-Jun-2026 04:36:49 JST Johnny Peligro

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • pistolero
     @7666 @phnt @p @lanodan @sun @lumi i ask chatgpt about this
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Wednesday, 24-Jun-2026 04:50:35 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Blurry Moon
     • pistolero
     @7666 @p @lanodan @sun @lumi
     >idk why this line of arguing still happens.
     
     For the same reasons why there are people that refuse to use Brave, only because Brendan Eich is the CEO; refuse to have functional hardware capable of anything more than having one browser tab open, because imaginary "freedom". We live in a world where everyone's opinions are all around us and coming up with our own opinions takes time and thought. But inheriting opinions of people you like/talk to around you is cheap and quick. It creates kind of herd mentality and orbiters around certain opinions, with way less people trying to form their own opinions based on their own decisions and thinking processes.
     
     Ask most of these people the reasoning behind their opinions and you'll either get silence or "I just don't like XYZ". Why don't you like XYZ? Crickets. Ask most people that don't like Trump now why that is, and few will be able to answer that without going on one of the routes that they will never explain, like "he's fascist/a nazi". Few will give you their own reasons instead of some herd opinion from their circle.
     
     In a way we live in a post-thought reality.
     
     >https://infrablog.lain.la/ai-crap
     >rsync
     
     Give this a read: https://medium.com/@tridge60/rsync-and-outrage-d9849599e5a0
     pistolero likes this.
     pistolero repeated this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 04:58:14 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     @phnt @7666 @lanodan @lumi @sun
     
     > For the same reasons why there are people that refuse to use Brave, only because Brendan Eich is the CEO
     
     I cannot get it to compile without downloading a binary last time I checked.
     
     > refuse to have functional hardware capable of anything more than having one browser tab open, because imaginary "freedom".
     
     Well, I do disagree with this part. I can have a shit-ton of browser tabs open and all of the things that want me to download a binary are shitty things. (Have had to use Macs for work. The things are unusable and Apple hates to let you tweak shit.) You have it, you hack it, but I try not to have things that make it harder to hack. I'm happiest if I've got a healthy /usr/src and anything the machine is doing that I don't want it to do, I can gut.
     
     > It creates kind of herd mentality and orbiters around certain opinions
     
     We have always been this way and it's actually a survival trait.
     
     It is frustrating that people go with whatever is working right now instead of whoever has been consistently correct, but that is from the perspective of an individual. If you consider the emergent properties of this on a village with a population near Dunbar's number, ten thousand years ago, then everyone makes jokes about the guy with OCD being obsessed with washing his hands (a lot of mental disorders have constant rates across populations and cultural barriers; OCD, psychopathy, ass burgers, and the numbers are generally 1-3%, so most Dunbar number populations will have 1-3 guys with one of those disorders) until there's a plague and his house is the only one that isn't dying of plague and then the herd (or hivemind if you prefer) starts listening to that guy, despite the fact that he's largely superstitious. What he does gets incorporated into weird rituals that solve the problem and prevent its recurrence and then when the plague is gone he gets ignored again.
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Wednesday, 24-Jun-2026 05:11:56 JST 7666

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     @phnt @p @lanodan @sun @lumi The issue is that his shipping velocity went through the roof without the same level of review and care taken. We can imply this because regressions were had, and it was easy to put two and two together (see images), and I saw no justification for them other than "I tried my best bro". Rsync is a mission critical tool for many, many people and businesses (although that whole "the entire world depending on one dude doing it for free" image comes to mind) and that level of carelessness is core to my argument that effort is still required for AI tools.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Wednesday, 24-Jun-2026 05:11:56 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Blurry Moon
     • pistolero
     @7666 @p @lanodan @sun @lumi
     >I tried my best bro". Rsync is a mission critical tool for many, many people and businesses (although that whole "the entire world depending on one dude doing it for free" image comes to mind) and that level of carelessness is core to my argument that effort is still required for AI tools.
     
     On the other hand, he owes you absolutely nothing. He wants to do other things than maintaining rsync with basically zero outside help from anyone for years, so you are left with two choices. Get an LLM involved to get at least some level of maintenance and security fixes going, or remain in previous status quo and do almost nothing. These are your only two options you have. Your third choice is to go somewhere else.
     pistolero likes this.
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Wednesday, 24-Jun-2026 05:12:12 JST 7666

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     @phnt @p @lanodan @sun @lumi I mean, I owe absolutely nothing to everyone who uses my services, but at least some level of moral framework keeps me from packing up shop unannounced or injecting AI slop into everything that I do.
     
     Why does being a reasonable person have to be formalized?
     pistolero likes this.
     Nietzschean Ekko Enjoyer repeated this.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Wednesday, 24-Jun-2026 05:12:38 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Blurry Moon
     • pistolero
     @7666 @p @lanodan @sun @lumi I wouldn't call rsync slop yet. Neither 3.4.3 or 3.4.4 failed me yet and I use rsync extensively.
     
     >Why does being a reasonable person have to be formalized?
     
     Doesn't has to be. His actions to me so far are reasonable. If he decides to stop development of rsync completely, I might be annoyed for some time, but I can't really blame him. What is not reasonable to me on the other hand is the Mastodon™ users having a meltdown over Claude in commit attributions universally all day every day, and seemingly running newest versions of software in production without any testing, as I've seen during the time where the rsync regressions cropped up. Reasonable people would make an issue on Github about it and maybe write a post about the regression along with pasting the link to the issue, but that is not what happened in 99% of the cases.
     pistolero likes this.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Wednesday, 24-Jun-2026 05:13:18 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • pistolero
     @7666 @lanodan @lumi @p @sun Here's a thought, how do you think people would react if the regressions were introduced completely with any AI use at all.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 05:13:49 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     @phnt @7666 @lanodan @lumi @sun Like the libxz compromise?
   • Embed this notice
     Nietzschean Ekko Enjoyer (r000t@ligma.pro)'s status on Wednesday, 24-Jun-2026 05:27:38 JST Nietzschean Ekko Enjoyer

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • pistolero

     @phnt
     Hating everything AI has become part of the Online Goodthink meta, so yes this means that Canceldon and Plebbit users are going to bitch about it literally every chance they get.

     What's super funny about this is that some leftists who use AI have convinced themselves that *liking* AI is the correct leftist progressive position, and that hating it is Magat Trumper behavior, and then they turn around and get dogpiled and nazibar'd by their own team
     @7666 @p @lanodan @sun @lumi

     pistolero likes this.
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Wednesday, 24-Jun-2026 05:28:58 JST 7666

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     @p @phnt @lanodan @sun @lumi That's the primary point I made in my big post thing. AI provides the perfect lens to magnify all the shitty tendencies of people, and (as one such example) roving bands of Indians who see a cash cow because they can now speak like a Westerner via AI re-interpreting their "do the needfuls" are EVERYWHERE and this means the amount of AI output that ends up being junk is in the high 90% range because the person behind the wheel is exceeding their skill level or not acting in good faith. This means the 1-3% of outputs that aren't junk is usually also coming from people putting *effort* into what they are doing to achieve something useful and have enough background knowledge to know how technology is supposed to work, and those people will succeed.
     
     But the perception of care and effort is hard to show when AI is involved because I really just don't think the vast majority of people (especially those in management positions) understand what it is they're dealing with and don't care to, they just want the funny chat prompt to shit them out more profit/code/grifts/efficiency/etc. The biggest issue I currently have "scaling AI" is abstracting how to use it for other people to *not* pull that shit on me (e.g. skills or plugins or whatever), and then doing all the regular software development things like version control management, lifecycle management, documentation, deployment, etc. that AI isn't great at. Ironically these newbies shitting out slop also consequently suck at all those things because, guess what, it requires effort.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 05:28:58 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     @7666 @lanodan @lumi @phnt @sun
     
     > the 1-3% of outputs that aren't junk
     
     Well, here's something interesting: experienced devs got a productivity increase from LLMs; junior devs saw a *decrease*. Treating it like something that can think for you will absolutely fuck you, and it is nearly guaranteed that people will do this and then blame the tool when it fails to do the thing that it isn't good at doing and then the vendors will try to make it do the thing that people wish it did.
     
     > I really just don't think the vast majority of people (especially those in management positions) understand what it is they're dealing with
     
     Yes, accurate.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 05:33:17 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     @7666 @phnt @lanodan @lumi @sun
     
     > Lopping off support for Linux <5.6
     
     Holy fucking shit *why*?
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Wednesday, 24-Jun-2026 05:33:18 JST 7666

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     @phnt @p @lanodan @sun @lumi I don't know that it's relevant. In both cases, code regressions are caused by a lack of rigor in the development process. I think it's more of a lightning rod if that lack of rigor was due to AI but still, buggy software gets ridiculed no matter who pulled the trigger on the latest commit and the trust in that software will be damaged.
     
     Also the regressions are stupid. Absolute paths breaking? Seriously?
     
     https://github.com/RsyncProject/rsync/issues/922
     
     And this is exactly the kind of stuff that would get missed, making assumptions AI is very good at. Lopping off support for Linux <5.6 would be a pretty big change, but it was just kinda slipped in there.
     
     https://github.com/RsyncProject/rsync/issues/905
     pistolero likes this.
   • Embed this notice
     Nietzschean Ekko Enjoyer (r000t@ligma.pro)'s status on Wednesday, 24-Jun-2026 05:34:17 JST Nietzschean Ekko Enjoyer

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • pistolero

     @phnt
     Suddenly stop talking about it and conveniently never correct anybody who brings it up as an example in the future
     @7666 @p @lanodan @sun @lumi

     pistolero likes this.
   • Embed this notice
     Nietzschean Ekko Enjoyer (r000t@ligma.pro)'s status on Wednesday, 24-Jun-2026 05:52:25 JST Nietzschean Ekko Enjoyer

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • pistolero
     • lolitechengineer

     @lolitechengineer
     Have you considered inverting that entirely?
     @lanodan @phnt @7666 @lumi @sun @p

     pistolero likes this.
   • Embed this notice
     lolitechengineer@loli.church's status on Wednesday, 24-Jun-2026 05:52:26 JST lolitechengineer

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Nietzschean Ekko Enjoyer
     • 7666
     • Phantasm
     • Blurry Moon
     • pistolero
     @r000t@ligma.pro @phnt@fluffytail.org @7666@comp.lain.la @p@fsebugoutzone.org @lanodan@queer.hacktivis.me @sun@shitposter.world @lumi@snug.moe I personally like to go by the much simpler philosophy of:
     "If it's doing something good, it's good"
     "If it's doing something bad, it's bad"
     
     No reason to overthink it, nor does it have to be a permanent title nor an absolute in all applications.
     pistolero likes this.
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Wednesday, 24-Jun-2026 06:25:41 JST 7666

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     @p @phnt @lanodan @sun @lumi somebody did this without checking
     
     #include <linux/openat2.h>
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 06:27:00 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     @7666 @lanodan @lumi @phnt @sun Seems like rsync should stick to userspace headers instead of trying to bother the kernel directly.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Wednesday, 24-Jun-2026 06:31:11 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Blurry Moon
     • pistolero
     @p @7666 @lanodan @sun @lumi xz-utils was a case of a malicious actor taking over a project from someone that didn't want to maintain it. Not an AI introducing dumb regressions. See above reply for what I think would be the case instead.
     
     Although they do share similarities, like completely vanishing from online discourse within a few days.
     
     Also I think it's safe to say that I meant "without any AI use at all". I love dyslexia and not reading what I wrote :)
     sshd-backdoor-xz-jia-tan-oreally.jpg
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 06:33:47 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     @phnt @7666 @lanodan @lumi @sun
     
     > xz-utils was a case of a malicious actor taking over a project from someone that didn't want to maintain it. Not an AI introducing dumb regressions.
     
     That was my point, yes. You asked how people would feel if it was done by means other than an LLM. (Technically you asked "AI" and LLMs are ML but are not AI, as they predict tokens but do not form abstract knowledge representations, but I don't want to derail with pedantry.)
     
     > I love dyslexia and not reading what I wrote :)
     
     I did read what you wrote.
   • Embed this notice
     lolitechengineer@loli.church's status on Wednesday, 24-Jun-2026 06:38:32 JST lolitechengineer

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Nietzschean Ekko Enjoyer
     • 7666
     • Phantasm
     • Blurry Moon
     • pistolero
     @r000t@ligma.pro @lanodan@queer.hacktivis.me @phnt@fluffytail.org @7666@comp.lain.la @lumi@snug.moe @sun@shitposter.world @p@fsebugoutzone.org but if I inverted that, it would create conclusions that violate objective truth.
     For example "cute anime girls are good" is objectively true and fits with my current worldview, but inverting it to viewing them as bad is objectively untrue.
     pistolero likes this.
   • Embed this notice
     SilverDeth (silverdeth@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 06:48:00 JST SilverDeth

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • pistolero
     @p @7666 @lanodan @lumi @phnt @sun
     
     At some point, you start to wonder if this is intentional sabotage by bad-faith actors.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 06:48:30 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • SilverDeth
     @SilverDeth @7666 @lanodan @lumi @phnt @sun I'm actually kinda past wondering, starting around the time Congress called it irresponsible to release uncensored models.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Wednesday, 24-Jun-2026 06:50:45 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Blurry Moon
     • pistolero
     @7666 @p @lanodan @sun @lumi
     
     >I don't know that it's relevant. In both cases, code regressions are caused by a lack of rigor in the development process. I think it's more of a lightning rod if that lack of rigor was due to AI but still, buggy software gets ridiculed no matter who pulled the trigger on the latest commit and the trust in that software will be damaged.
     
     It's relevant, because I don't think the fuss would be that big. Instead I think you would see more apologetic posts that mention that he's maintaining it alone, is almost 60 and wants to do other things, things like that.
     
     >Also the regressions are stupid. Absolute paths breaking? Seriously?
     
     Yes, I know. It also shows that rsync is/was kinda abandoned. Maintained projects of this size and importance usually have tests that would catch these completely dumb bugs, but rsync just doesn't and the test suite is basically fully LLMed.
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 07:04:13 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     @phnt @7666 @lanodan @lumi @sun
     
     > That is mostly a part of a problem I call "normies using computers".
     
     I am interested in ways of preventing this problem from persisting.
     
     > They simply don't care.
     
     Something something "The spread of computers and the internet will put jobs in two categories: people who tell computers what to do, and people who are told by computers what to do." but with standard "Andreessen sucks" disclaimer.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Wednesday, 24-Jun-2026 07:04:14 JST Phantasm

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Blurry Moon
     • pistolero
     @7666 @p @lanodan @sun @lumi
     >But the perception of care and effort is hard to show when AI is involved because I really just don't think the vast majority of people (especially those in management positions) understand what it is they're dealing with and don't care to, they just want the funny chat prompt to shit them out more profit/code/grifts/efficiency/etc.
     
     That is mostly a part of a problem I call "normies using computers". A tool in the hands of someone that doesn't know how to use it. And it's not like these people will learn when you tell them that you can't fully trust what comes out of the chatbot with barely cited sources. They simply don't care. Humanity had this problem for thousands of years, but misusing those tools usually had tangible and immediate consequences like cutting your fingers off on a saw, or burning yourself while forging with glowing metal. LLMs so far don't have those quick and tangible consequences that force people to learn.
   • Embed this notice
     Växẍ Säbbäth (vaxxsabbath@detroitriotcity.com)'s status on Wednesday, 24-Jun-2026 07:08:22 JST Växẍ Säbbäth

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • pistolero
     • SilverDeth

     @SilverDeth @7666 @phnt @p @lanodan @sun @lumi it’s mostly structural, embedded in the design of the LLM itself (I will not elaborate)

     but yeah, the interesting thing is that AI is going to lead to a circulation of elites; before, if you squinted hard, the relationship between managers/executives was that of user+AI - the manager/executive came up with “the vision”, entered it into their console (line managers), who then orchestrated the tasks of subagents to make sure it got done and returned the result, which was then refined by further vision

     there are only so many seats for those managerial/executive positions, however, and those seats have increasingly been reserved for the most retarded and self-serving idiots known to Mankind, which is the genesis of basically all of our current problems

     AI threatens to hand one of those seats to anybody who can afford a subscription to a sufficiently powerful model and has the talent to guide it to make the right things and clean up the resulting output, which is one of the reasons we’re seeing those subscriptions tick up in price; the retard-oligarch egregore is dimly aware of the threat posed by structural barriers falling and creating the possibility of actual meritocratic competition for their wealth and social status (“haha, jackass, you got your executive role by being a jewish nepobaby, I got mine by having a genius idea and having AI implement it for me, we are not the same”)

     ultimately, they are going to attempt to gate AI (you see this now with model lockdowns and compute scarcity) - they do NOT want this capability “out there in the world”, because the thing it is most dangerous to is their own positions; they are demonstrably useless and are acutely aware of that fact, and they know for a certainty that if they are ever forced into a competition connected to real outcomes they are going to lose and very badly

     whether they succeed in locking it up, however, remains to be seen

   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 07:08:22 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • SilverDeth
     • Växẍ Säbbäth
     @VaxxSabbath @SilverDeth @7666 @lanodan @lumi @phnt @sun
     
     > which is one of the reasons we’re seeing those subscriptions tick up in price; the retard-oligarch egregore is dimly aware of the threat posed by structural barriers falling
     
     No, this is not social engineering, this is normal VC mechanics. They try to spend each other into the ground in order to corner the market. It has gotten prohibitively expensive to subsidize so now they're solidifying positions. OpenAI is racing to get an IPO ahead of their MSFT bills coming due so that they can avoid an acquisition, because they've hit the ceiling on market share.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 07:11:24 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Nietzschean Ekko Enjoyer
     • 7666
     • Phantasm
     • Blurry Moon
     • lolitechengineer
     @lolitechengineer @r000t @lanodan @phnt @7666 @lumi @sun
     neuralink.png
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Wednesday, 24-Jun-2026 12:42:09 JST 7666

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • Phantasm
     • Blurry Moon
     • pistolero
     • SilverDeth
     • Växẍ Säbbäth
     @p @VaxxSabbath @phnt @SilverDeth @lanodan @sun @lumi >charge so much for compute that the bill would equal the value of the company so you just acquire them and write off the charges which are essentially bullshit anyway
     pistolero likes this.
   • Embed this notice
     pistolero (p@fsebugoutzone.org)'s status on Wednesday, 24-Jun-2026 13:08:40 JST pistolero

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • SilverDeth
     • Växẍ Säbbäth
     @7666 @SilverDeth @VaxxSabbath @lanodan @lumi @phnt @sun $250b minimum buy over six years. Probably headroom was in the front and the first bill's likely due 2027Q1, given that they're targeting a 2026Q4 IPO.
   • Embed this notice
     :marseyloadingneon: m0xEE :marseyloading: (m0xee@breloma.m0xee.net)'s status on Tuesday, 07-Jul-2026 17:49:21 JST :marseyloadingneon: m0xEE :marseyloading:

     in reply to

     • Haelwenn /элвэн/ :triskell:
     • 7666
     • Phantasm
     • Blurry Moon
     • pistolero
     @7666 @phnt @p @lanodan @sun @lumi
     Seems like they've got it fixed eventually still — according to the linked issue.
     But I agree, things like this just slipping in, and some people justifying it post-factum based on completely unrelated things ("OMG, you are using no longer supported kernel — just get new hardware, lel") is utter insanity :marseysigh:
     pistolero likes this.