Arborium is AI slopware and should not be trusted
50 points by FedericoSchonborn
50 points by FedericoSchonborn
Amos accused zkat (the maker of the miette error-handling Rust library) of “witch hunting” after zkat added Amos to a list called “open slopware”, which I have heard was meant to list software and developers that use AI. Perhaps in part due to Amos’ comments (I am not entirely aware of the timetable), “open slopware” got a lot of backlash, which led to zkat receiving a deluge of harassment
if you put people on a list with a public derogatory name, i don't think it's fair to describe the backlash as "harassment" (unless is crosses a line. i don't feel like digging in further, but i haven't seen that from the posts directly linked from this article). that's picking a fight with a bunch of people at once, which understandably gets you a lot of negative comments at once
and, a complete lack of self-awareness to then put up a blog post harassing this amos guy, for having the audacity to publish software that doesn't work well
I wish we had an unkind flag for articles. This amounts to a complaint that a newly released library isn’t bug free and perfectly documented. Yes it’s fair to point that out and say it’s not ready for prime time, but this hectoring tone is basically bullying a project for admitting they use coding agents.
that's perfectly acceptable behaviour on lobste.rs as it's part of the wider llm punching-down. obviously this would be completely unacceptable behaviour in any non-vibecoding related domain
Wait, are you saying that the anti-AI position is "punching down"?
(Note, it's not anti AI, it's anti LLM being used to produce code). Well as it gets expressed on lobste.rs, definitely. It's punching down because there are a large group of people who are not the technical experts on lobste.rs, they are people perhaps without the skills to be professional developers who don't have deep experience or domain knowledge to e.g. write algorithms or perfect interfaces, write beautiful code, develop good architectural designs, do perfect TDD, or learn Lisp for the fun of it, understand deep beautiful concepts of technology and get to practice them from a place of 'craft'. And those people say completely authentically: 'This technology is amazing for me. It is enabling, it is letting me doing things I otherwise can not do, in ways that are revolutionary to me and I am excited. I am producing creative, functional work in a new way and it is having positive benefit for me, my clients, my life'. And the domain experts at lobste.rs say basically one thing: 'that is slop - a term we invented specifically to describe the thing you are producing which has no value. You are doing it wrong, we own the domain of 'programming' and doing valid work in the domain looks how we say it looks. Not only is there no value or benefit to your particular use of this new technology, there is a negative value because it makes us look at low quality code when we don't want to and massively increases the quantity of what we call 'bad code' in our world. It's not our problem you don't know that's bad code and you should not be producing things that way'. The situation is complicated because of people using LLM's to do things like prepare and submit low-effort low-value PR's to open source projects or the use of LLM's and automation to attack other people, but these are uses of technology by bad actors; it's complicated because of the economic interactions and power dynamics in tech which are getting discussed in parallel. This is complicate because these are social problems, they are about the boundaries we have around our communities and systems. It's these social boundaries and assumptions and conventions that are getting rapidly challenged, destroyed and haphazardly rebuilt by the effects of the new technology, but the technology itself is getting the blame, rather than how it's being applied and our collective ability to deal with social and political challenges.
When someone in any domain with more experience and knowledge berates the work, time or effort or someone who has less, that is punching down. Not only is it poor behaviour, but it's shameful behaviour because the next generation of programmers are going to rapidly develop in this new world and we are disconnecting them from the current old guard where a huge body of knowledge and expertise is held.
The old guard is facilitating this disconnection, not finding out how their domain knowledge can be applied there.
What is being done here is shitting on somebody's project (clearly with an axe to grind, looking at the zkat stuff) in the name of being anti-AI. That's not punching up for sure, and it's disingenuous to call this just being anti-AI.
this is a hit piece. i hope to see better articles on the lobster's frontpage!
the author ran into issues because they used deno instead of node, and choose this opportunity to be angry about the maintainers LLM usage. this is open-source and nobody owes anybody anything.
ironically, the author of the post themselves state (on the issue) that they would be unwilling to put in the effort to setup the project locally with the fix and test things out. but willing to put in the effort to write a hit piece however.
You’re given steps on how to use Arborium in simple use cases, but you’re completely fucked the moment you step outside the golden path
heh, well, in fairness I don't think that's unique to AI slop :-D
Come to think of it, if all this mythical "llm gonna take our jobs" was remotely true, it would be absolutely painless to include in the "generate me a website" prompt the extra "and document all the use cases" instruction since the whole exercise is touted as "just ask and it will do it, eventually"
To be fair, the blog post's main issues as presented seem to boil down to trying to execute the library in a non-browser environment, which Arborium doesn't claim to support anywhere. A completely reasonable response from an OSS maintainer perspective would've been to respond "we don't support Deno, PRs welcome".
Of course, the underlying point of the post is to raise suspicion whether the entire library is vibe-coded and as such untrustworthy but I feel that the arguments raised are strawmen and thus insufficient against the claim.
main issues as presented seem to boil down to trying to execute the library in a non-browser environment,
Web is not my expertise, but that's how it looks to me as well. Maybe I'm missing some detail, but there's nothing there that would made me doubt the quality of the software or point at the fault of LLM being used, so whole article comes across as just baseless shaming.
The description above is accurate. The one thing I think actually reflects (somewhat) poorly on the project is that Amos opened and landed a PR which claimed to make "arboriumHost work in Node.js, Deno, and workers", which fixed only the immediate bug and did not actually make it work on Node or Deno. This would have been obvious if it was tested according to the test plan in the PR, but the test plan is in fact not checked off, so I assume it simply wasn't.
But hey, I've also landed PRs which fixed a simple bug which was masking another bug and not done the testing necessary to reveal the second bug. It's not like this is some major sin, especially given that the project is (even after this PR) not documented to support running in Node or Deno at all.
The description says "so arboriumHost works in Node.js ...". Not sure if "arboriumHost" is "the whole project". But it's splitting hair, I guess.
The whole project just needs a proper test suite for running in: Deno, Node, etc. before it can officially support them. That's just how working with LLMs is. And realistically, thanks to LLMs it's much easier to roll out and maintain extensive test suites like that, which somewhat pays for their "probabilistic" nature.
Generally there's a "skill" to using LLMs, and big part of it is anticipating failure modes, and creating rails to keep it productive and reliable. The code can "write itself" to a certain degree, but now the effort moves into making sure it's not defective, either via manual review, or building more checks and balances.
Oh oh no, oh no. I've participated a little bit in the slopware list drama referred in the blog, and had a brief, utterly unsatisfying exchange with Amos, over on Mastodon. It seemed odd to me how defensive they've gotten, given nobody has at that point (at least to my knowledge) accused them of vibecoding any of the projects they've been so passionately posting about, merely of "using LLMs at all".
https://hachyderm.io/@hjvt/115864492695758983 Archived: http://archive.today/HmVHy
Your link doesn't work. "Not Found (yet?)" Works now and new link was provided.
I only looked at the https://hachyderm.io/@fasterthanlime/115861122803393499, and I think Amos is right. Preparing some kind of lists shaming people is a witch hunting like. And then people implying that his content is suddenly bad because he uses LLM is just deranged and rude.
The whole crowd posting under everything "this look like LLM touched it", etc. in recent months has already been very common and annoying. Now it looks like people are compiling some shame-lists. And blog-posts shaming you that your library didn't work, and that must be because you vibe-coded it. Maybe soon they will be calling employers etc.
I don't agree with the approach of that list, but I do share many of the ethical concerns that motivated its creation.
A lot of tech influencer types have been very reckless with promoting LLM use, and could stand to experience a little shame for that.
The problem I see is that we get a plenty of people sharing the concerns, compiling lists, writing exposé posts and whatnot but no one actually going and writing a lib that highlights a hundred languages using proper parsers.
I'm missing the link between those except that this particular instance is about a syntax highlighter. What is the problem?
The problem is that people are eager to criticise but not so eager to do the work, even as badly the the work they criticise.
Here is the list, resurrected by the power of forking:
https://codeberg.org/small-hack/open-slopware
I don’t condone nor condemn. I am simply posting it as the original was removed and it’s always good to have as original sources as possible available in a debate.
I mean, they were placed on a list called "open-slopware". Even if the specific claim against them is more measured ("Core developer openly uses LLMs for library dev"), the list name itself is obviously meant to be pejorative, no?
Getting defensive doesn't seem odd at all to me.
so, I get the fundamental frustration here. I've spend quite a bit of time as of late running into projects where features just made no sense at best or were nonfunctional at worst, because the maintainers just YOLO'd vibecoded slop in without even bothering with a cursory examination.
but...I'm not convinced any of the issues raised here rise up to that bar? "oops I used window on code that I forgot is used outside the browser" is quite literally a thing I've done on my own before, and "I had to use undocumented options to make things work"...I mean, that's just the average library experience. really, the sum of the entire article is an experience better than a sizable amount of libraries I've been stuck using!
I can't really see how this becomes "and cannot be trusted". that seems rather steep, no?
i don't really know how to feel about this post.
on one hand, more i am more and more feeling the disenchantment of vibecoding and vibe-coded apps (for anything larger than a 1-5 file python script/util or similar, which are basically one-shottable), as well as using vibe-coded stuff being unpleasant as a user.
on the other hand, was this post necessary? last time i checked, foss library authors don't owe the user anything. they clearly prioritised speed over correctenss, possibly as a means-to-an-end scratching-their-own-itch type of sideproject. sure, the library may be "bad" and missing docs, so just use a different one, contribute fixes or write your own solution (like the author did).
i'm failing to see the need to publicly shame a library its author.
foss library authors don't owe the user anything
Legally? No. (Well, to the extent that such disclaimers have legal force anyway.)
Socially? We live in a society, and have to live with the outcomes of each others' actions.
And speaking of Arborium specifically, it's not just uwu small irrelevant library with a git repo; there's a polished-seeming website trying to sell it to you and Amos is campaigning to get it adopted by docs.rs.
Warning people of it makes total sense to me. Do you also get mad at your bank's warnings about scam callers? The scammers don't owe you anything either!
You're forgetting that "person i liked used thing i think is bad so that person is bad and needs to be shunned" is in play here. I think it's dumb, but whatever it's the author's life.
Another case of slop derangement syndrome? Crypto is so over, now hating on LLM usage is what cool kids do.
So a library didn't work in deno, and of course it's a proof that it was all LLMs fault. So lets write a blogpost accusing the whole code for being shit because author is using LLMs.
I just wrote a disclosure recently and I'm adding it in all the READMEs, predicting that some people will be having visceral reactions. Might as well get it out of the way and not have to deal with them at all.
Leaving aside the hitlist I can see the authors disappointment that one of their respected FOSS people is in reality a Vibecoder, even more so when it happens on a project with that much visibility and PR (blog post, lobsters front page etc).
Never meet your heroes and maybe never look up how their coding practices look under the hood, you might find emacsAI ;)