I've contributed 32 edits (1 new page) in the past 10 years, so despite being stable, there are still many things to add and fix!
Sadly, the edit volume will likely drop as LLMs are now the preferred source for technical Linux info/everything...
I've contributed 32 edits (1 new page) in the past 10 years, so despite being stable, there are still many things to add and fix!
Sadly, the edit volume will likely drop as LLMs are now the preferred source for technical Linux info/everything...
At the same time, I suspect resources like the Arch Wiki are largely responsible for how good AI is at fixing this kind of stuff. So I'm hoping that somehow people realize this and can continue contributing good human-written content (in general).
> So I'm hoping that somehow people realize this and can continue contributing good human-written content (in general).
AI walled-gardens break the feedback loop: authors seeing view-counts and seeing "[Solved] thank you!" messages helps morale.
Definitely, being unpaid LLM trainer for big corporations while nobody actually reads your work is not very encouraging. I wonder what the future will bring.
I do think we will, at some point, face a knowledge crisis because nobody will be willing to upload the new knowledge to the internet.
Then the LLM companies will notice, and they’ll start to create their own updated private training data.
But that may be a new centralization of knowledge which was already the case before the internet. I wonder if we are going to some sort of equilibrium between LLMs and the web or if we are going towards some sort of centralization / decentralization cycles.
I also have some hope that LLMs will annihilate the commercial web of "generic" content and that may bring back the old web where the point was the human behind the content (be it a web page or a discussion). But that what I’d like, not a forecast.
I wouldn't be surprised if LLM companies end up sponsoring certain platforms / news sites, in exchange for being able to use their content of course.
THe problem with LLMs is that a single token (or even a single book) isn't really worth that much. It's not like human writing, where we'll pay far more for "Harry Potter" and "The Art of Computer Programming" than some romance trash with three reads on Kindle.
This is perhaps true from the "language model" point of view, but surely from the "knowledge" point of view an LLM is prioritising a few "correct" data sources?
I wonder about this a lot when I ask LLMs niche technical questions. Often there is only one canonical source of truth. Surely it's somehow internally prioritising the official documentation? Or is it querying the documentation in the background and inserting it into the context window?
LLM companies already do this. Both Reddit and Stack Overflow turned to shit (but much more profitable shit) when they sold their archives to the AI companies for lots of money.
I kind of fear the same. At the same time I wonder if structured information will gain usefulness. Something like man pages are already a great resource for humans, but at same time could be used for autocompletion and for LLMs. Maybe not in the current format but in the same vein.
But longer form tutorials or even books with background might suffer more. I wonder how big the market of nice books on IT topics will be in the future. A wiki is probably in the worst place. It will not be changed with the MR like man pages could be and you do not get the same reward compared to publishing a book.
> nobody will be willing to upload the new knowledge to the internet
I think there will be differences based on how centralized the repository of knowledge is. Even if textbooks and wikis largely die out, I imagine individuals such as myself will continue to keep brief topic specific "cookbook" style collections for purely personal benefit. There's no reason to be averse to publishing such things to github or the like and LLMs are fantastic at indexing and integrating disparate data sources.
Historically sorting through 10k different personal diaries for relevant entries would have been prohibitive but it seems to me that is no longer the case.
Absolutely. Even though I don’t use arch (btw), the wiki is still a fantastic configuration reference for many packages: systemd, acpi, sensors, networkmanager I’ve used it for fairly recently.
You see it referenced everywhere as a fantastic documentation source. I’d love seeing that if I were a contributor
Also if it's not correct someone else will edit it. But with the LLM it's just the LLM and you, and if you correct it is not like it will automatically be updated for all the users.
I just installed Arch (EndeavourOS) and LLM did not help. The problems were new and the LLM’s answers were out-of-date. I wasted about 5 hours. Arch’s wiki and EndeavourOS’s forums were much better. YMMV
They may be preferred, but in a lot of cases they’re pretty terrible.
I had a bit of a heated debate with ChatGPT about the best way to restore a broken strange mdadm setup. It was very confidently wrong, and battled its point until I posted terminal output.
Sometimes I feel it’s learnt from the more belligerent side of OSS maintenance!
Why would you bother arguing with an LLM? If you know the answer, just walk away and have a better day. It is not like it will learn from your interaction.
The Gell-Mann effect? If you can't trust LLM to assist with troubleshooting in the domain one is very familiar (mdadm), then why trust it in another that one is less familiar such as zfs or k8s?
Maybe GP knew the proposed solution couldn't have worked, without knowing the actual solution?
Arguing with an LLM is silly because you’re dealing with two adversarial effects at once:
- As the context window grows the LLM will become less intelligent [1] - Once your conversation takes a bad turn, you have effectively “poisoned” the context window, and are asking an algorithm to predict the likely continuation of text that is itself incorrect [2]. (It emulating the “belligerent side of OSS maintenance” is probably quite true!)
If you detect or suspect misunderstanding from an LLM, it is almost always best to remove the inaccuracies and try again. (You could, for example, ask your question again in a new chat, but include your terminal output + clarifications to get ahead of the misunderstanding, similar to how you might ask a fresh Stack Overflow question).
(It’s also a lot less fun to argue with an LLM, because there’s no audience like there is in the comments section with which to validate your rhetorical superiority!)
1 - https://news.ycombinator.com/item?id=44564248 2 - https://news.ycombinator.com/item?id=43991256
> It was very confidently wrong, and battled its point
The "good" news is a lot of newer LLMs are grovelling, obsequious yes-men.
I think it all comes down to curiosity, and I dare think that that's one of the main reasons why someone will be using Arch instead of the plethora of other distros.
Now, granted, I don't usually ask an LLM for help whenever I have an issue, so I may be missing something, but to me, the workflow is "I have an issue. What do I do?", and you get an answer: "do this". Maybe if you just want stuff to work well enough out of the box while minimizing time doing research, you'll just pick something other than Arch in the first place and be on your merry way.
For me, typically, I just want to fix an annoyance rather than a showstopping problem. And, for that, the Arch Wiki has a tremendous value. I'll look up the subject, and then go read the related pages. This will more often than not open my eyes to different possibilities I hadn't thought about, sometimes even for unrelated things.
As an example, I was looking something up about my mouse the other day and ended up reading about thermal management on my new-to-me ThinkPad (never had one before).
Depends on how AI-pilled you are. I set Claude loose on my terminal and just have it fix shit for me. My python versions got all tuckered and it did it instead of me having to fuck around with that noise.
I'm not there yet. Not on my work system anyway.
Seen too many batshit answers from chatgpt when I know the answer but don't remember the exact command.