1.4.0 is still in the "oven", full of bugs and mostly developed by "robobun".
I suspect that the diarrhea that the AI shits out actually made the project unmaintainable.
time will tell, but being 3 months without a release for bun doesn't seem good, based on their release schedule so far.
What do you mean telling an AI to one shot convert a project from one language to another is a terrible idea, I LITERALLY told it to make no mistakes when prompting it.
Bun was bought by anthropic and forced to become another one of their shitty "Look goys, AI can code gooder now!" just to dump the project, just like their rust C compiler that was worse in every way to GCC even after having everything done for them and it didn't even work at the end to do what they set out to do.
Can AI code, mostly, can it generate good code, no, and the more you make it generate the worst off it gets.
Macs suck. Even art fags only use the fucking things because they're "the default" at their faggy art schools and ad agency gigs. Trying to write software or wrangle devops shit on a Mac is an exercise in anus tonguing. At my last job, they insisted I use a Macbook instead of a vastly-better-spec'ed PC I had available (even after offering to wipe it and let them install their preferred OS on it to make it "compliant" with corporate requirements), and six months after I started when the layoffs came, they were shocked into silence when they gave me the "bad news" and my reaction was "oh thank fucking Christ I can ship this piece of shit Macbook back to you guys. You might as well send the remote wipe command right now, I'll take this thing to FedEx the minute we're off this call."
Great idea if you want your journalctl logs to be spread out over thousands of different directories (since logs get written to /var/log/journal/<machine-id>), along with other quirks I can't predict because systemd assumes /etc/machine-id is constant. I assume none of you actually tried this but updooted anyway because you like to stick it to the man.
You're really not doing systemd the big favor you think you are by pointing out yet more ways it fucks up the simplest fucking things so consistently, and in the worst conceivable fucking way at every fucking opportunity.
"detect-fash"? Is that a joke? Detects "problematic" configurations? And they rejected it?
Interesting. I maintain a very old open source project (its first public release was in the late 90's), pure C, that begrudgingly started shipping a systemd unit file a few years ago (probably because some asshole submitted one and it didn't break when my maintainer at the time tested it). About six months ago some smartass posted a PR proposing a broad swath of changes to it, adding all sorts of ridiculous fine-grained restrictions on it to "lock it down" because some automated systemd-adjacent (or systemd-provided) utility analyzed the current unit file and declared it to have an arbitrary score of 10 out of 100, and so I'm bad and I should feel bad.
Meanwhile the actual software in question? All it actually needs is to be able to listen to a network socket (optionally -- it can operate locally only without ever touching the network) and talk to a hardware port of certain kinds (it varies based on the hardware you're driving with it). That's it. It doesn't need (or want) root, it doesn't need a socket under 1024, it doesn't even read files besides (again, optionally, its config file; it can be configured entirely on the CLI if desired). It's small and lightweight enough to run on ESP32's (and does in at least two known production environments), and as such even modernizing the code to C17 requires careful review to make sure we don't kill support for microcontroller environments.
But here comes a 50+ line change (almost all additions and no removals) listing all the shit it doesn't try to do anyway, just to make some draconian systemd-furnished tool stop whining about it being the most insecure daemon ever written for computers. The damned thing's been around for thirty fucking years, has only ever had one CVE, was popular enough to be audited several times by very enthusiastic "assholes on purpose" types (at our grateful invitation), and is about as simple as it can get, but nope, it doesn't conform to the "new" systemd way (which I don't recall ever being consulted on) and so some random tool whines and leads to useless PRs. And you just know the minute I merge those fucking changes some random n00b is going to try to set this thing up because it looks neat and fucking systemd is going to give him no end of shit trying to start up this software and he'll instantly assume it's because it's "old and crusty" and not systemd being a cunt like it always is.
I think so too and have said so before. You point current gen AI at a large codebase in an obscure language and have it one-shot translating it into another obscure language.
I think this i close to unmaintainable by humans at that point. It is vibe coding now gping forward, for new features and bug fixes. It is 100% vibe coding going forward.
Even if it was human maintainable. Imagine trying to maintain a huge pile of machine translated shit from one fag language to another fag language? I would move to canada and apply for MAID if I was that human.
What do you mean telling an AI to one shot convert a project from one language to another is a terrible idea, I LITERALLY told it to make no mistakes when prompting it.
Bun was bought by anthropic and forced to become another one of their shitty "Look goys, AI can code gooder now!" just to dump the project, just like their rust C compiler that was worse in every way to GCC even after having everything done for them and it didn't even work at the end to do what they set out to do.
Can AI code, mostly, can it generate good code, no, and the more you make it generate the worst off it gets.
People seem to think that the frontier labs actually employ skilled developers, which is an absolute joke if you look into what they produce. They could've gotten good results in the experiments you mention if they actually employed skilled people.
The most recent joke is the "sandboxing" that they're panicking over the the LLMs "escaping", and when you look into it they broke every security rule of thumb possible. But even older stuff like the "piss filter" because openai didn't seem to know about the imagenet normalization constants or forgot to update them, or the hilarious bugs in the CLIP code they released, etc.
They employ a bunch of mediocre ea rats, that don't understand the bitter lesson (and why it invalidates their RSI hype), playing with toys (the attention mechanism, CoT, etc.) developed elsewhere by smarter people, and a ton of funding.
You're really not doing systemd the big favor you think you are by pointing out yet more ways it fucks up the simplest fucking things so consistently, and in the worst conceivable fucking way at every fucking opportunity.
I'm not trying to do it any favors, I dislike systemd. I was describing what happens in it (and in every other DHCP client with default settings). I even mentioned how /etc/machine-id is blatantly unnecessary.
When I think about all those developers - not just at AI labs but also FAGMAN generally - getting paid half a million US$ to occasionally turn out buggy jank into a world that has no alternative, the only conclusion I can make is
When I think about all those developers - not just at AI labs but also FAGMAN generally - getting paid half a million US$ to occasionally turn out buggy jank into a world that has no alternative, the only conclusion I can make is
Funny joke, but FAGMAN employs a ton of insanely skilled people, e. g. the amount of human brilliance wasted on eeking out another 0.5% on some webshit benchmark in chrome is insane. Or the people Microsoft employs to work on low-level ML stuff like onnx-runtime seem very skilled (I'm skipping tensorflow because it seems to have been a shitty and slow middle management ploy to own the stack by throwing man-hours at it, not something cooked up and owned by engineers with a good idea).
The frontier LLM labs seem to be uniquely populated by technical midwits and have endless amounts of money.
People seem to think that the frontier labs actually employ skilled developers, which is an absolute joke if you look into what they produce. They could've gotten good results in the experiments you mention if they actually employed skilled people.
The most recent joke is the "sandboxing" that they're panicking over the the LLMs "escaping", and when you look into it they broke every security rule of thumb possible. But even older stuff like the "piss filter" because openai didn't seem to know about the imagenet normalization constants or forgot to update them, or the hilarious bugs in the CLIP code they released, etc.
They employ a bunch of mediocre ea rats, that don't understand the bitter lesson (and why it invalidates their RSI hype), playing with toys (the attention mechanism, CoT, etc.) developed elsewhere by smarter people, and a ton of funding.
From my understanding, this is kind of a feature of the environment they've built. Any older/wiser and more security conscious developers don't eat up the marketing hype from these labs, and are probably filtered out before they apply or early on because they're not as enthusiastic in their application.
Anecdote: I applied to anthropic when I saw a job ad for them. I didn't think I'd get it, since it was a few days after it had been posted, but I work in a relevant area. The application process asked lots of attitude questions towards the business and AI before getting you to fill in your experiences and give your resume. I can imagine this process causing less enthusiastic people to not bother. That coupled with what was likely an insane amount of applications probably means that they'd ignore a lot of the later applicants who don't spend their time on social media and actually do good work.
Anecdote: I applied to anthropic when I saw a job ad for them. I didn't think I'd get it, since it was a few days after it had been posted, but I work in a relevant area. The application process asked lots of attitude questions towards the business and AI before getting you to fill in your experiences and give your resume. I can imagine this process causing less enthusiastic people to not bother. That coupled with what was likely an insane amount of applications probably means that they'd ignore a lot of the later applicants who don't spend their time on social media and actually do good work.
A lot of larger c*mpanies are concerned more with this faggoty invisible social marketing game than they are about producing quality goods, every employee needs to be a tow-the-line sucky sucky 5 dollar PR goylem first and a productive asset second. Attitude is the currency of choice so 'culture fit' supersedes 'craft quality' any day of the week, especially if you work for one of these zombie corporations that live off of hype and investor funding a la AI frontiersmen. Even past that, everyone is obsessed with image, it is the gayest shit imaginable and it trickles both up and down through the pozloaded negholes of both big corpo drones and individuals at any tier of business be it self owned or employed by another.
Speaking of *BSD, has anyone else noticed how almost everyone on YouTube who talks about OpenBSD is just a poser? For example: https://youtube.com/watch?v=_RO9beApnecMy assumption is that it's due to OpenBSD being seen as "The 1337 haxor OPSEC demon OS". I have never seen this issue with FreeBSD.
Most FreeBSD users seem to be legit oldfags who use it for some specific reason. NetBSD seems to be somewhat similar. Like I'd use FreeBSD from time to time as a single purpose device. It's also really well documented. Even when I was new to it (although not BSD Unix in general) back in the late '90s, I could set up a router/firewall with a simple HOWTO.
I used this because Windows at this period of time was such a joke in terms of security you couldn't even have it plugged into your DSL directly without being infected by some bullshit before the install was even finished.
