I want to preface this by saying I am not anti-AI/LLM, especially in the way it has manifested on social media. In fact, I’m highly critical of the movement due to how abusive the movement has gotten (especially on Bluesky and the FOSS community) and dismissive to information that contradicts their claims, and I do not trust them for reliable information on this topic and consider them very manipulative. I also avoid giving into tribalistic behaviours because I consider them hazardous for my ADHD brain, which is why I think I’ve taken pragmatic solutions more than ideological ones. Even as I’m critical of LLMs in this post for cognitive reasons I want to make this clear, and if you don’t like that, feel free to close the tab. Sorry not sorry lol.
I’m not “pro-AI” either, as this post will explain. I’m in the boring middle, and I just wanna share my experience for people who happen to be in my predicament.
I’m not really using any large language model at the moment, because I found a very real problem that kept occurring: overthinking loops around decisions that LLMs kept fuelling, especially in programming.
It’s no secret I have decision paralysis up the ass, and with programming, there are many ways to implement the process you want. In this specific instance, I wanted a way to store objects of varying types in C#, for my music library organiser that I want as a modern drop-in for MusicBee. I wanted to store custom tags that had different types (string, numbers, ratings, etc.) in them, and this led me down a path into LLMs. I was asking Claude about this and we devised an abstract class system for it, and eventually it chose to do a factory design pattern for the track fields since one can have fields that display different data (e.g. MusicBee can show “year” and “year (YYYY)”), which went right over my head and stressed me out. That was a signal for me to stop, and I haven’t really touched the code since.
I was asking people in programming communities how to learn certain C#/object oriented programming paradigms or how to decide what is best, however what I should’ve been asking was how to stop striving for perfectionism, and how to deal with my ADHD while coding. And with that, I decided to unsubscribe from Claude and think about it as I played Tomodachi Life: Living the Dream, which does a better job at getting non-artists to pick up the pencil than anti-AI has ever done in my opinion.
For me, to solve any decision making problems, I’m going to abuse the KISS principle to start off with. Keep things as simple as they can be, and only when more complex behaviour is truly needed I will start adding that on. MusicBee does a lot of simple things, too – for the “year (YYYY)” field, it doesn’t use DateTime at all, it just substrings the first four letters. Does this have parsing issues? Yes! Does it matter in the real world? No. For those using full dates in the Year field when tagging their music library, just tag your music using the ISO standard (YYYY-MM-DD) and there you go. A reverse engineering effort of the MusicBee library format also showed that MusicBee saves tags with a number (likely denoting a tag’s enum) and then a string value – my app will probably do something similar.
That’s not to say LLMs don’t have their uses. LLMs are great at grabbing a lot of web results quickly, and summarising things for you to evaluate of your own accord. I found Espanso, a text replacer application, because I asked an LLM for an equivalent of AutoHotkey that would work on macOS. LLMs are also great at figuring out where my trail of thought stops, which I’m having major problems with because of my ADHD. I don’t know how many times an LLM caught me stopping something mid-coding or mid-sentence and it flagging this. I’ve yet to figure out a solution for this particular problem that works for me, that doesn’t involve a language model. (And I’m all ears on this! I’m asking fellow programmers diagnosed with ADHD or anxiety in particular!)
Even for basic shit, they can be good, or even fine at worst. However, a decision maker or even decision helper they are not, and I’m happy I realised that with my own experience.
There’s a lot of mental processes I’m trying to curb this – learning to actually trust my instinct is one of these procedures. Learning what the Traveller’s Dilemma was also helped me with decision making, helping me curb perfectionism and making me look towards Hugo again. The decision to use WordPress was a result of LLM use I’ll have to admit, and I think it’s partly that reason I’m looking at messing with Hugo on a separate site, and allowing myself the room to experiment. I’ve also made the decision to avoid Astro entirely, because of the simple reason of not wanting to learn or even vibe-code JavaScript at all. If I go static, I want to avoid anything JavaScript, especially Node.js. I am also a very risk-averse guy, and I worry a lot that my decisions are a risk that will blow up in my face – this is a core tenet of decision paralysis that’s prevalent with ADHD brains, and it’s why it took me so long (outside of finances) to actually pull the trigger on buying a Mac.
For some people, all this may be obvious. Hell, it feels like in hindsight I should’ve stopped the moment I went back to Windows. However, some people – especially those with ADHD or any other reasons for a busy, overwhelmed mind – may not realise it until something hits them and they’re like “wait, hang on a second”, like what happened to me. There is value in saying things no matter how obvious they are, because for some it might not actually be. Despite my opinion that we’ve hit the top of the Sigmoid curve on the progress with this tech, LLMs are still relatively new, and we’re all still figuring out how to deal with it.