So I’m curious. I read in Arch is for experienced users topic that one should choose a distro on the basis of what you’re going to use it for. I can only think of one I would recommend (and that’s tentatively), which is Ubuntu Studio if and only if you are into music production. But it’s just a configuration of Ubuntu.
In all the distros I’ve ever used I’ve only used them for some programming, some GIS work (QGIS and Grass), LaTeX, git, playing a synthesiser or two from my midi keyboard, zoom, browsing the web, shopping.
At work I needed a real-time Linux for a robot control system, which allowed me to choose which cores to run my code on exclusively i.e. deterministic real-time. So I suppose that’s an example.
Am I missing out? Does it matter which Linux I use? (And if it matters, does it matter that it matters? – Douglas Adams)
For desktop use, almost every mainstream distro will run what you listed. The differences are mainly: package freshness, release cadence, and how much you want to babysit it.
None of that is about what you use it for, it’s about how you like to maintain the system.
You picked the distro for robot control so yes there are such cases.
For security, pentest and network admin work you technically can build a Kali equivalent on Arch or Debian, but you’d spend a weekend doing what Kali ships preconfigured. Same logic for REMnux on the malware analysis side.
Servers with a hard support contract. RHEL or Ubuntu LTS, because someone is paying for the SLA, not because the kernel is different.
Anything where the hardware vendor only ships one set of drivers. NVIDIA Jetson, certain industrial boards, some SBCs. The distro is whatever the BSP supports.
Older or very low-RAM hardware. Not really a “use case” but a constraint that narrows choices.
So I’d say for 90% of what most of us do, distro choice is aesthetic and habitual. Whether that matters or that it doesn’t matter, I’ll leave to Adams.
I’d add to the differences above, a distribution’s stance on proprietary licenses and software. Some distributions (e.g., Debian, Fedora) draw a harder line inclusion or not of non-free software. There’s always ways around it of course, but using non-free software such as drivers is easier on distros such as Ubuntu or Linux Mint.
That is question of means and purposes. It matters when certain Linux distribution fits one’s purposes best. It does not matter until one is subject to a corporate policy or other kind of enforced discipline.
HOWEVER … that statement is incomplete and therefore misleading!
You must
first choose what you plan to do,
identify the tools with preferred GUI interractions,
and only then
choose the Distro(s) on which those tools are verified functional in the way you want them to be!
Otherwise, the consequence is likely Distro-hopping … in search of things working the way you want!
The specific domain of “Real-time Linux” is one where I question the use of the word “Linux” itself, since the Official Kernel is not coded in a structure dictated by real-time requirements.
For real-time, you quite literally need to “strip out” and “replace” key process structures to correctly work as real-time, which is scanning multiple input channels, simultaneously, transforming/processing the values simultaneously, then acting based on priority in a defined-sequence within an absolute-limit timed cycle.
That is not how “standard Linux” is coded or operates.
If you don’t try different distros you will never know whether they suit your needs.
I disagree. At least in the last 5-10 years or so, the capabilities of different distros have converged, even if the look-and-feel hasn’t. Provided the distro has friendly support, I think a newcomer could choose any they want. For a long while at work, I used Scientific Linux (CERN variant of Red Hat) and following that CentOS before it got discarded. With those I found support was good but very formal. I used CentOS as a central server for postgreSQL. So some Linux’s were (still are?) very business oriented without so much going on as the latest popular distros.
I think that Arch can be used for a lot. EndeavourOS is a good choice if you want to learn Arch without the installation, but you won’t have dracut out of the box on vanilla Arch. I have my setup optimised for heavy tasks. I have the CachyOS kernel and repos on my manually installed Arch.
In my (not so) humble opinion, I think that Arch isn’t too difficult to install. You just need to read the wiki. I personally believe that the hardest part of the installation is configuring the bootloader. But I use only Limine (and systemd-boot when the CachyOS kernel breaks and I need to use the normal one), so I can’t really compare.
Totally agree. I’ve seen posts in other forums that Arch is not for beginners but I don’t know the reason why they say that. When I installed “Ubuntu Studio” on my second drive, it merrily reinstalled Grub despite the website saying it supported systemd booting. (I got rid of Ubuntu in the end.)
Furthermore, a person’s experience with a distro depends more on the user front end, i.e. the display manager, the windows manager and the desktop environment than the underlying OS itself. (Admittedly, there is some overlap and certain dependencies across these three, and I am glossing over some inconvenient details here.)
Arch (and no doubt most other distros) allow you to make those choices after installation or during. But like most distros, you can just stay with the vanilla theme and “standard” choices until you want to explore a bit. Customising is great fun. But Arch Linux is still Linux, about the only CLI difference across distros is the command to run the package installer. Be it yay, pacman, yum, or apt. There really is not much to learn for a beginner that isn’t common to all Linux varieties.
For me, the two most amazing things about Linux are its stability (yawn), and the way you can customise everything. You’ll soon have rainbow coloured folder icons, whizzy login screens and splash screens, and menus where you want 'em. Much of all that available on-line from arty folk who specialise in creating themes. It’s all so friggin easy.
Regarding Arch Linux - or any distribution for that matter, nothing is truly difficult for anyone with a fifth grade reading level plus the ability to carefully follow directions.
I can vaguely remember at least thirty years ago when I first tried out Arch Linux. It was not difficult at all to install, but it was necessary to either follow a set of steps or create your own set of steps to install whatever you were interested in and then manually configure any programs or packages that may be automatically set up on some distributions. On the original Arch Linux nothing was configured for you. It wasn’t difficult, but there were details for each installation. The positive was that you could build and then have exactly what you wanted. The cost was that it was all up to the person doing the work.
Endeavour OS and Cachy OS are two of the modern variations based on Arch Linux. Both are very easy to install and ready to use immediately after installation. Both of them offer different interfaces and tools, and in those collections, it’s easy to install and configure additional software with very little incremental effort.
I’m partial to Endeavour IS myself; I like the appearance and the tools are just right. Though I’m mostly a user of Debian-based software, for several years now I’ve added Endeavour OS to my software collection of distributions because it has been a really good addition!
From my personal perspective, I have noticed that Linux in general is something new and the digital language can be quite confusing.
However, it is 2026 and there are a plethora of resources that anyone can use or have access to, to teach themselves or help with learning. On top of that with the assistance of AI you can make tasked that would take years to figure out only take minutes. Even if you are one of those that do not support the use of AI, if you simply search for the problems, I am sure someone has run into the same or similar issues and have a possible solution to your problem.
So whether you are a “Purist” or a “Vibe Coder” you can learn what you need with some time and really just a little bit of effort.
Yes indeed: between the expertise of different people cooperating and offering assistance and the increasing power of systems, which now allow the creation of quickly indexed, search, and locate, the addition of this information, as the reliability improves, this can become “Intelligence” of a type, and we label a lot of it AI.
The concepts of Artificial Intelligence (AI) are not new at all. In fact, when I was quite young and just beginning my studies in Computer Science, my professor, who was my teacher in ALMOST all of my Computer Science courses because our program at the time was fairly new, it turned out that his doctorate studies at the University of Iowa was, believe it or not, in Artificial Intelligence. AI was NOT a buzz word back then, it was an idea that was hoped to eventually be feasible. Well, it now is. What gets me is that not too many people know that kind of history, but all of the big companies are trying to remake themselves, claiming all of this stuff is life changing and fabulous.
That’s where we all have to be careful. Yes, absolutely, there is outstanding potential to do a lot of things quicker and more effectively. The danger is the same as it always is and always has been. Does anyone know what GIGO means? I’ll tell you in case you do not. GI - Garbage In, GO - Garbage Out.
Also, movie science fiction fanatics absolutely HAVE to remember the movies of the past - one of my favorites was The Terminator. This comes with an exaggeration with a warning - not handled right, robots, machines, computer “intelligence”, and so forth can indeed be helpful, but you also need people like Sarah Connor to give birth to people like John Connor to “tame” and keep advances in technology “honest”. So that’s what I say now as AI has become a reality. It really can be helpful, but it can also be destructive, so those who build it and those who use it should be responsible to do it with care and reason; then it can be extremely helpful.
Don’t think for a moment that “the bad guys” aren’t already using it for “bad” and for criminal activities.
Even between 1995 and 2000, the huge company Microsoft did a study and even created their own “What’s possible” technology film and sent it to many other technology companies. I saw it in the nineties when I worked at Digital Equipment Corporation.
They showed how it would, very soon, be possible to use remote devices, such as laptops or personal communication devices, to do things away from the office - maybe even from a park or a beach - BUT - the bad guys at the very same time would be intending to do criminal things with the same technology.
This is quite germane to the AI discussion today. Technology, in and of itself, is neutral. It can be used for nothing, it can be used for specific things, it can be used to process information at record speeds and that also means that it enables things to be stolen and compromised, also at record speeds.
So here’s the take away: it doesn’t mean we can’t use this stuff, but for every powerful technology we create, it’s important to protect it, provide a trace of what’s going on, so in my opinion for every new “super power” you create, you have to harness that super power appropriately.
I hope that just as much is written about these aspects as is written about all the hype around each of them. As that is done, perhaps the comfort levels can relax; without the right checks and balances there is a significant and real reason to be quite concerned.
That last line is what a lot of people skip over. Whether you’re coming at it as a purist who wants to understand every layer or someone who just wants to get something working, the resources are there either way. The Linux community specifically is pretty good about that, someone has almost always hit the same wall before you and wrote it up somewhere.
The AI assisted learning thing is real too. I’ve used it to work through stuff that would have taken me days of digging through man pages and forum threads. The key is you still have to understand what it’s telling you, you can’t just blindly paste output and call it done, especially on Linux where a wrong command can actually hurt you.
The GIGO point is one of those things that sounds simple but actually cuts pretty deep when you think about it in the context of where AI is right now. Everyone is focused on the output, the speed, the novelty of it, but almost nobody is asking hard questions about what the models were trained on, who curated it, what got left out, and what biases got baked in so early that nobody even noticed. A model that’s confidently wrong at scale is genuinely more dangerous than one that’s obviously broken, because people trust the confident answer and stop verifying.
The history piece is something I wish more people knew. AI as a serious academic discipline goes back to the 50s, Turing, McCarthy, Minsky, people who were thinking about machine intelligence before most of the hardware existed to even test the ideas. It hit walls every decade or so, the AI winters, when the promise outran what compute and data could actually deliver. What changed around 2017 and exploded after that wasn’t a new idea, it was transformers, massive datasets, and enough GPU power to finally make the theory work at scale. The concept is 70 years old. The hype makes it sound like it appeared three years ago.
That Microsoft film you mentioned is a detail I hadn’t heard before and it’s a good one, because it shows that the dual use problem was understood internally by the people building this stuff long before the public conversation caught up. Every powerful technology follows the same arc. The internet was going to democratize information and it did, and it also became the infrastructure for fraud, surveillance, radicalization, and state level influence operations. GPS was going to revolutionize navigation and it did, and it also made precision weapons cheaper and tracking people trivial. The pattern is consistent. The difference with AI is the speed of the transition from controlled to ubiquitous, and how general purpose it is compared to any previous technology. A GPS chip does one thing. A capable language model can write code, generate disinformation, craft a phishing email, explain a vulnerability, or impersonate someone, all from the same weights.
The “technology is neutral” framing is technically true but I think it lets builders off the hook a little too easily. A hammer is neutral. A system optimized at massive scale to produce convincing output regardless of whether that output is true is a little harder to call neutral, because the optimization target itself is a design choice someone made. The neutrality argument works better for the underlying math than it does for the product decisions around how these systems get deployed, what guardrails get built in, and who has access to what.
Your point about checks, balances, and traceability is the one that actually matters most in practice. The scariest version of this isn’t Terminator, it’s systems making consequential decisions at a speed and scale where the audit trail is either nonexistent or too complex for anyone to meaningfully review. That’s not science fiction, that’s already happening in credit, hiring, content moderation, and parts of the legal system. The answer isn’t to stop building, it’s to be serious about the logging, the oversight, and the accountability structures, which right now are lagging badly behind the deployment curve.
The people who built nuclear technology understood the stakes and built entire frameworks around containment, inspection, and treaties before broad deployment. We have not done anything close to that with AI and we’re already well into the deployment phase. That gap is the real concern, not the technology itself.
