Things I wish I learned when I started using GNU/Linux (for noobs)

Chat about Zorin
1

First, there is an app for almost everything.

Usually, there are multiple competing apps for one task. If you use GPT or Claude or Grok, they'll all very commonly give you a much more complicated solution than what's really needed to accomplish something. They'll immediately tell you to go to the terminal, install a bunch of things you've never heard of, start nanoing into system config files, etc. But 99% of the time, there is an app, somewhere, that does all of this for you, usually with a convenient UI.

A good example is recently I wanted to remap an extra button on my mouse to be recognized by the custom keyboard shortcut option in Gnome settings. GPT told me to install keyd and jump through a thousand hoops of terminal commands to remap it. However, it was entirely unnecessary and didn't even work. I just went into the Software app included with Zorin, and got Input Remapper. A couple minutes later, problem solved. No terminal needed.

Second, there is an extension for basically anything else.

Extensions in Gnome are amazing. Zorin heavily utilizes this system to deliver the experience. Extension Manager (from Software) is an essential app. The Alphabetical App Grid extension solved my only real beef with the Gnome desktop by automatically organizing all of my apps. Other extensions like GJS OSK made the touch screen keyboard on my laptop just the way I wanted it.

Third, Unison.

This is more for people who don't like using cloud storage, but can be equally useful even if you do. Unison is a cross platform app I've had since my Windows days. It is without a doubt the most useful app I've ever had for keeping backups without needing to manually manage every file that changed. Basically it goes through two directories and figures out how to make them identical. The first scan is always slow, but after that, it's FAST. Like, ludicrously fast. I keep a LUKS encrypted SSD USB on me always as opposed to using any cloud based services to back up files. If I lose my PC, I have the USB, if I lose my USB, I have my PC. It's not a bulletproof solution, but nothing self hosted ever truly is. I keep organized notes on each distro on it. If I hit the same problem a second time, I have notes on it, or usually just a shell script I have GPT type up based on what worked.

Even just making a small dedicated folder on your PC for your solutions and keeping them backed up will save you time and energy in the future.

Lastly, AI.

AI, for all their problems, are the reason there are more Linux users now than ever before. That's a great thing. However, be aware that no matter how advanced they get, they will eventually mess something up if you blindly copy and paste their instructions into the terminal. Linux is great because it assumes you know what you're doing. But Linux is also fragile for that same reason. Windows assumes you don't even know what the terminal is. So the AI habits one might acquire in using it on Windows will not often translate well to Linux. Ask GPT for citations, check its sources, and question it. Make sure the recommendations it's giving you are necessary and valid for your situation.

Where AI thrives is automation. Say you just solved a problem and you like how it worked out. Save the commands you typed, and give them to GPT to give you a shell script to run in the future. If you happen to do some distro hopping, these scripts will be valuable assets if you end up back with Debian again. Zorin, Ubuntu, and Mint are all based fundamentally on Debian, and the problems you get (or more commonly create for yourself) on one will more often than not happen again on the next.

That's all. Hope this helps someone. If not, it was still fun to type out, lol

6 Likes
2

Glad to hear stories like this. Just moved this to Chat as it seems more apt there versus a tutorial.

2 Likes
3

Which AI do you find to be most useful?

I've had the best luck with Grok. ChatGPT had me install Fuse2 to run AppImages in Ubuntu, which then erased the Gnome DE. I've tested other AIs with this question to see if they get it right. Grok was the only one I've used thus far that realizes you need to install a Fuse compatibility layer to make AppImages work that are designed for Fuse2.

1 Like
4

I would not use Grok on principal.
Perplexity A.I. is good.
The key element to receiving a good A.I. response is to ensure you create a good extensive question that covers all the salient points or you end up with G.I.G.O.

1 Like
5

Don't worry, Grok will get something else wrong, later.

The way LLM's work is through pattern synthesis and that is based on a pretty simple "Majority rules" premise. I will give you an example.
Recently, testing an A.I. (Granted, a trained and filtered model, but through Jan), I mentioned the shame in that we lost Ozzy Osbourne and it corrected me to point out that Ozzy is very much alive.

What happened, looking through the tokens - decades of comments about a living Ozzy outweighed the number of comments on a non-living Ozzy. Statistically, it judged Ozzy must be alive. And it lacked the capacity to think it through, understand outliers in data sets or... Check the answer.

A.I. must be taken with a lot of salt. It is fine for refining ideas, exploring underlying scientific ideas, but not good for fact checking, current events or... Coding.

Some A.I. is built explicitly for code building and those are the ones that GnuLinux users should look to - and even then, with a whole lot of salt.

A live person knows what an outlier is - or relies on real world experience to differentiate between patterns.
A.I. structures an image that covers the base foundation and has the illusion of an answer, but just enough nonsense to turn a coding problem or troubleshooting into gibberish, no matter how confidently it asserts it.

We are actually working toward a plan to start limiting how A.I. is used on this forum.

Because until now, it seemed to be harmless or just another resource.
However, as time has passed, more and more we are seeing a pattern emerge that is more than just risky - I hope we can mitigate these effects before tragedy.

5 Likes
6

I agree. I've found AI to be most useful when it's treated like an advanced, interactive encyclopedia.

I have, however, found it to be useful for Linux troubleshooting. For example, I wanted to run music editing software and use a bunch of plugins I purchased for Windows many years ago. Grok helped me set up a ton of background processes that were unique to my particular configuration, hardware, software, and accessories. It took about two hours in the terminal, but I had a decently working audio system.

Also, since I run Ubuntu, there's a ton of forum information that it can sift through to find answers to common questions that I'm not readily seeing in the top search results (and sometimes that's because I don't know the proper terminology to get the search results I need).

That said, there have been countless times that it tells me to do something that doesn't work because it ignores the fact that I'm running a newer version of the distro, kernel, or DE than what it has in its training data. It'll hallucinate extensions for Gnome, assume packages are available that aren't, etc. Generally, if I don't understand what it's telling me to do, I don't do it.

Another example: I played around with RHEL for a day or two just for the fun of it. AI was generally very unhelpful with that, and probably with anything that's a bit more niche.

1 Like
7

Exactly, you hit the nail on the head.
Here's the bigger problem: These propagate on the web at large, adding this information to the training data as "fact" for the A.I. to continue to statistically favor.
This is the thing we need to confront.

2 Likes
8

Or Skynet.

1 Like
9

The good news is... That is still well in the realm of fiction at current levels and tech. :wink:

3 Likes
10

It's not the AI I fear in a Skynet-like scenario, it's the billionaires developing them for things they should never be used for (i.e. automated weapons systems, self driving cars, self flying planes and helicopters, job replacement, surveillance, etc.). Governments and corporations have placed far too much confidence in its capabilities after seeing some pretty sentences regurgitated by a chat bot. A modern day Skynet would most likely be a problem of stupidity rather than one of competence.

The main unavoidable downside I see with it currently is that we will no longer be able to trust that any video is genuine. Many people will be framed for things they never did in the coming years, as Grok so kindly demonstrated. But the uses will continue inevitably towards international psychological warfare and political disruption.

Sorry, getting very off topic now, I just like ranting about this stuff lol.

3 Likes
11

Six xAI Co-Founders Exit Amid Grok Controversy, SpaceX Merger

https://www.eweek.com/news/xai-founder-departures-musk-reorganization/
This was posted today and told me everything I needed to know about Grok.

1 Like
12

I agree, I use Perplexity AI. It's very good and also will give you a break down of the sites it used to find the information. Which is helpful, as you can then double check that it's not just made something up. Find that Chat GPT has it's moment with this.

1 Like
13

That is the rub with AI search. If you can't trust it and have to read articles anyway what the point of using AI for search? At least in it's current state.

3 Likes
14

I am not specifically using AI for search, I just notice that often the AI summary at the top of search results in many search engines is flat-out 180 incorrect. Most of my searches are on specifically technical questions to do with various programming questions, but I've noticed it on more general queries too. If I was to follow that generated answer I would be down the wrong path immediately.

1 Like
15

Would you perhaps thinking of documenting this in a tutorial? My studio still runs on my older Win10 PC. I have a mix of hardware and software. I have some GForce instruments which I love, but the biggest hard-thing-to-lost would be NI Reaktor which is an awesome (and infuriating with its non-updated framework) sound creation tool.

I just installed Ubuntu Studio on my old (13 years old) laptop to test it out, and am pleasantly surprised how well that went. it can't run much of VCV rack but...

i also attempted Zorin18 on it. The basic install went well, and Z18 seems to work with no hitches, which is encouraging on this older machine. I followed some tutorials on installing and wiring up Jack, ALSA etc and became hopelessly lost, hence the trial of Ubuntu Studio which is preconfigured. I'd prefer the elegance of Zorin.

My plan is to get as much working on the old laptop to demonstrate proof of concept, then consider building a new box to replace the Win10 machine.

if you are willing to relate your experiences and steps, I am sure this would be of interest to many who are looking at moving from Windows. I can't imagine running a studio on Win11 with its updates and interruptions.

Mods - this might need a new thread? Thanks!

1 Like
16

It's a powertool.

The point for me is the speed and context AI offers. It is an answer machine much more powerful and potentially useful than a mere search engine. Of course I don't just trust the answers are correct - same thing with search engines. I want to see where the info came from, (the guiding principal of Perplexity.ai's creation.)

And I can't have a "conversation" with a search engine. There is no "remembering" the context. With each search I am starting over, refining my question for more applicable results, putting quotes around certain phrases to avoid all the non-applicable results that keep appearing, resorting to pasting and copying the question from a text editor because it is becoming so unwieldy to retype, etc.

When I have the answer, if in doubt I'll run it by the forum. Before I pull the trigger, I'll back up.

Also I'm getting better at asking followup questions.

"Of the solutions you show here, which one has proven to succeed the most often?"

"You say [quote goes here], but earlier you said [earlier quote] - please explain."

"Ok, so say I do what you suggest...what are the downsides?"

...stuff like that.

Why continue to use a hammer when I have a nail gun at my disposal?

1 Like
17

Treat AI as... a slightly better google alternative. Partly this is due to the worsening of google as a service over the years (from their own desire to show ads to the whole concept of SEO which shifted what gets presented first), and partly due to LLMs getting a bit better.

Still, if getting stuff done with AI is your thing, make sure to give it the correct prompting instructions. Typically LLMs are great at giving a list of recommendations, but they're really bad at giving instructions - especially if those instructions get interrupted due to an error at any point in the process. It's thus worth changing the prompt so that the AI only gives you one command at a time so you can verify the results before you continue.

There's one downside to the above which is... query limits. If you're already paying for some kind of AI service then this probably doesn't apply, but for anyone not paying for AI (i.e. most people) it'll quickly "burn" through whatever is on offer.

Ex. using Gemini's slow model is much preferred to it's fast model... but if you start spamming it with one-command-at-a-time questions you'll quickly be forced to use the fast model.

1 Like