Ask Lemmy

Since I want to write in a bit more detail about my personal LLM/AI journey so far I will just give a Tl;Dr here: I started as a hater for that barely functional piece of junk that dared to call itself AI and slowly started to come around once it became better and more reliable. Now, let's start with the story time.

I still remember when cleverbot was the best way to talk to a computer and it was pretty shit, then suddenly ChatGPT popped up one day. I obv. had to give it a try and I was amazed that we already got that far and it happened basically over night (I know now that at that point the first LLMs existed for quite some time but it took a while for them to blow up and therefore me discovering their existence). At that point, they were pretty inacurate, couldn't even do simple math (which I found really ironic since that is basically the only thing a computer can do and very well too) and was inaccurate most of the time. I always had to double check each output and had to manually fix it pretty much every time, if it even was salvagable in the first place. But usually, I was better and faster at any task I could possibly give it so I became anti AI, started hating on the technogy and mocking people who used it.

Half a year later I finally decided to give it another try and it actually improved. Simple math was correct most of the time, it started to get most facts straight and some tasks like formatting text was either correct or just had some small mistakes I could easily fix myself. That was the turning point for me where I stated using it more often and hating it less.

Now, I actually like AI/LLMs. I use it almost daily for questions about pretty much anything that isn't about recent news (which is still very prone to errors) or about a nieche topic (same issue). The main reason why is because I stongly prefer getting taught new things with the chance to ask follow up questions or to clarify things I didn't underatand. I also use it to do simple tasks for me I don't want to do myself, like formatting texts in specific ways. And I also use image gen whenever I need an image, be it for my profile picture, banners or anything else, since I hate drawing stuff myself, do not have the money to commission stuff from actual artists and noone but AIs want to do it for free.

And as a developer, I of couse use AI coding tools. However, I have to say that I'm not a fan of agentic coding, better known as vibe coding to most non-developers, where you tell the AI what you want and it does all the coding for you. In my experience, the result sucks. Too many unneccesary comments, too many checks that usually just degrade the performance of the software and at some point, shit just starts breaking, the AI can't fix it and the codebase is too messy for me to read into it and fix it myself so I have to start over and do it myself anyway. For small stuff, like a simple script to inject into a website to make a table sortable or a cronjob that deletes every file in a temp folder that is older than a week, I have no issue with letting an AI vibecode that for me, then reading through the result to make sure it actually does what it should before using it. But for larger projects, I just give the AI access to my codebase and then chat with it when I'm stuck, unsure which approach fits best or give it smaller tasks to do for me and always check the output manually, never trust it fully. So after auto-completing words, then lines, then code snippets/chunks, I see coding LLMs as the next step that lets me generate even lager chunks, or even entire files, but with a more cautious approach. And it replaced the rubber duck for debugging.

One thing I almost never do is use AI to generate, enhance or even correct texts I'm writing. Take this comment for example, I bet I've made multiple mistakes and an AI could find better words and write everything way shorter while still getting the important bits right, but for some reason I do prefer to rather keep the human imperfections in it. Only for more.. professional settings, I sometimes rely on AI to do a bettrr job at writing than me, and most of the time, it delivers.

Now, what I am still concerned about is the future of AI. What I haven't mentioned so far is that I never paid for AIs and never will. If they are trained on stolen works where the creators of said works never received any compensation, neither will any AI company see anything from me. But as long as I can use it for free, I will continue doing so. However, I fear that this will soon change, since investor funding will not last forever and looking at the current hardware market and electricity costs, they will need a whole lot of money to keep going. And I'm pretty sure ads won't cut it this time so I'm really curious how this will end: Will the bubble burst and every large AI company will have to downscale massively to survive or will enough people decide to pay to keep them running the way they do right now? Once again, I will enjoy the benefits while they last and run local models once they try to start charging me, or just return to doing everything myself like before, but I will never support them financially.