AI's Most Underrated Use Case
TL;DR: Debugging Linux Errors
When AI and LLMs (or Large Language Models) first started gaining popularity back in the day, I was skeptical. In fact, although ChatGPT might be my most used website/app of all time right now, I'm still skeptical. As someone who uses it a lot, I'd like to think I've figured out it's pros and cons. If I had to boil it down to one sentence, I would say that LLMs are only as smart as the person using them.
Sure you could get good results if you just started typing random shit in there. However, more often than not, you're probably going to end up with garbage. The best way to get decent results out of an LLM is to structure your prompt right, and make sure you're not only giving it as much detail and context as you can, but also as clear instructions as you can give it. It's like working with a junior team member that isn't the brightest, but has the best memory out of anyone you've ever met.
Before we go into what I think LLMs are good at, let's talk about what you shouldn't be relying on. There's the obvious ones like medical advice and therapy. Therapy is more of a slippery slope. It's fine if you use it as a supplement to actual therapy. However, don't take it's word as gospel because at its core, it has been designed to be more empathetic towards you. I wouldn't be surprised if we start seeing a lot more self-centered and delusional people around us, as more people start using AI's to make themselves feel better about their actions and choices. Anyway, I digress. The no. 1 thing you should not be using AI for, is as a replacement for your own brain. ChatGPT, and comparable AI models, can be great thinkers, but they don't start off that way. The only time I've seen them think well, is when you train them how to think. You give them the exact steps of your own thought process, and get them to do the mental heavy lifting for you. That could work, but again, it takes quite a bit of effort to get it to that point.
Now, onto the good stuff. There's a lot AI is really really good at. Especially manual work that would usually take you a long time. Like research, discovery and anything that requires sifting through large volumes of data. One of my favourite things to do is to upload a large document that I need to read through, and just ask it to pull out stuff for me. For example, say your client sends you a 200 page manual to better understand their product. You can upload that with a prompt that says "use only the information you have from this document I have shared. Do not pull information from outside sources or from your own knowledge base unless specifically prompted to. Give me an executive summary, and tell me if there's anything in specific I need to know about X,Y,Z". Following that, you can ask follow up questions. Therein lies the true power of an LLM, follow up questions. I can't tell you how many times I've been on a forum for some odd research task or another and had the thought "okay I get this, but what if X". Now, the AI can go out and take what the original forum poster said, take what I said, and scour the interwebs to get me the answer I'm looking for.
This brings me to the best thing AI has done for me till date - make being a linux user way less painful. Non-linux users, you probably won't get the struggle. Linux is great, arguably better than MacOS and Windows. I've been a linux dabbler since 2009, when I installed Ubuntu 8.04 on my house PC, on a 384kbps bandwidth plan. It was like a breath of fresh air. However, until 2020, I hadn't fully committed. I dabbled in and out of linux, whenever I got bored of Windows or MacOS, but always found my way back on a Mac. Even in 2022 when I did go back, issues with i3 and sway eventually had be back on KDE or Gnome. I see you non-nerds dozing off, so let's go into a mini explainer side tangent.
Linux isn't like Mac or Windows in the sense that you can choose every part of it. Don't like how the user interface looks? Not only can you choose a whole new theme, you can also just choose a whole new foundation. Gnome and KDE are some of the most common, they look and feel familiar to non-linux users. However, the more control you want over your system, the deeper into the rabbit hole you're going to go. i3 and Sway are tiling window managers, you don't need to know what that means, you just need to know that it's super barebones and takes a lot of time to configure just the way you want. It won't really work for you out of the box.
Where was I? Right, AI. So, it's 2020 and I move back to Linux on my work machine, but because it's a work machine, i start off on KDE. It's stable. Most apps just work. I can take screenshots and all just fine. 2 months in, I'm bored again. I'm watching youtube videos about i3 and itching. So I decide to take the risk. I take the weekend, and take a sick day on Monday (sorry Pat). 3 days is how long it took me to configure i3 to a point that I was comfortable using it as my daily driver. Even then, it wasn't perfect. It wouldn't go to sleep when I closed the lid, screen recordings would always crash, some apps would open in weird window shapes and I wouldn't know how to move them around. Honestly it was so fun, but it wasn't the kind of experience you would tolerate when you have shit to do. So eventually I gave up and went back to KDE.
2022 I left GOODSTUPH, returned my Linux laptop (which I had formatted back to windows don't worry) and started at Chillchat. I bought a PC, since I would be working from home full time now, and I was faced with the decision once more. Guess what I did? That's right. Windows. ew. I know I know, Windows is hot flaming garbage. However, I needed something stable and I didn't know what my work at Chillchat would look like, so I had to make the decision I made.
Eventually I ended up buying a Macbook Air because I needed an on-the-go computer, so I could take the risk to install linux on my PC. I ended up going with KDE again, because I could take more of a risk but not that much of a risk. This story is getting longer now.
Fast forward to when I'm at Otterdev, still on KDE, but now no longer in a position where I need that much processing power so I'm fine to use my Macbook Air as my main machine. I strip out KDE, and install Hyprland. At the time, Hyprland was the new sexy toy, like i3 and Sway, but new and with nicer animations. I've been on Hyprland for well over 2 years now, and the first year, stuff still didn't work that well. You may be thinking, I thought this was a post about AI? Yes. I'm getting to it.
You see, back in the day when things weren't working that well on Linux, like say for example your screenshots sometimes just fail. Your options were either spent 4 - 5 hours googling it and trying to fix it, or install a different screenshot tool. Now? 2 minutes. "Hey chatgpt, when I take a screenshot on hyprland (arch) sometimes it fails. how to debug and what do I do". Immediately you get a full step by step guide on how to debug it, tailored to your specific use case. So you follow the steps, paste the errors you get into chatgpt, and in a few minutes you're up and running. What's better than that, is how you can tailor it do things on another level.
This afternoon, I was annoyed that my screenrecorder app wasn't working as well as I would like it to. With 12 minutes of back and forth, ChatGPT wrote me a script that pulls up a customisable menu for me to take the kind of screen recording I want with the settings I need, compress the video and store it in a specific folder with a specific timestamp.
I guess what I'm trying to say is that even someone like me, a huge nerd, would shy away from more complicated Linux setups because I had limited time and a low risk appetite to play around and tweak, and now with the help of LLMs and AI, I can build my workflows to suit my needs so perfectly, without having to spend the time and effort I would have had to in the past.