On the Topic of AI

This page contains some uncensored swearing. I also wrote half of it while pissed off, so take it with a grain of salt.

Generative Machine-Learning Models are a strange and wondrous technology, and are generally referred to (somewhat inaccurately) as 'Artificial Intelligence' for their (mis)perceived ability to perform many actions as a human might.

That shit is everywhere these days. Emphasis on 'Everywhere'. I mean, seriously. It's the biggest buzzword since the fucking internet was created.

The Truth

AI is here to stay. Even after the bubble pops, we have invested so much money into it that it has integrated into the tools that everyone and their mother uses.

The Good

Do not get me wrong, AI has a lot of cool capabilities. As the technology has advanced, it has been heavily improving in non-creative areas like pattern-recognition and code.

As such, I am certainly guilty of using it a bit. It has definitely found some errors and inconsistencies where I turned a tired eye and has definitely found NuGet libraries I was unaware of.

But the really cool thing is that the open source community has gotten their hands on the industry. Unlike the expensive SOTA (State-of-the-art) models Deepmind and GPT and all them corps are chasing, there's a flourishing community training and releasing thousands of open models on sites like HuggingFace.

What's more is that many of these open models are getting smaller and more efficient. And I'm not exaggerating. I can run a full-featured LLM with thinking like Gemma 4 E4B or Qwen 3.7B ON MY FUCKING PHONE! How cool is that shit?

Granted, the phone (iPhone 15 Pro Max) is new enough that its chip is optimized for LMs and vision, so YMMV.

The Bad

Regrettably, it is imperfect at more complex tasks. I used claude to migrate the site to ASP.NET from plain HTML, and I can tell you I almost went through more effort than it would've taken to manually do it just to modify it for my own uses.

Yeah, it can do good. But we as a civilization seem intent on using it for all the wrong reasons. Corpos seem to believe that GenAI is able to completely replace a human being in any context. Spoiler alert: of course it fucking can't. It's jack shit but a tool. A very powerful tool, of course, but a tool nonetheless. A tool that can easily hallucinate, make inconsistent output, and pretend it never happened (because it never had the power to know in the first place). When used properly (under supervision by a skilled human) it's great. But alas, technology exists only as a way to extract more value from the economy and environment.

Also, they really like to steal the work of artists so some lazy fucks can pretend that prompting makes them notable. (I really fucking hate image and video gen if you couldn't tell. Don't generate images, kids.)

Another big problem is how few guardrails it has, especially with how easy it is for kids to use it, for FREE. I myself fell down that rabbit hole, using early ChatGPT to outsource most of my creative manpower. You don't even need an account to try most of these, which is concerning, considering that most web-accessible closed-source LLMs like GPT are prohibitively expensive both monetarily and environmentally to operate (often, for bigger corps, it seems to act as a loss leader). But what's more damning is that said models are generally designed to reinforce the user's impulsive use of them, which children are especially vulnerable to.

Again, open and local models that don't do this are really cool.

What I think

Any John fuckin' Taxpayer can use an LLM to vibecode his shitass app into existence. But in order to make something good, it's got nothing on the indomitable Human Spirit. We, HOMO FUCKING SAPIENS, are fucking awesome.

If you're thinking about using AI, ask yourself this simple question first:

Given time, resources, and documentation, could you do it on your own?

If that answer is no, then don't bother. Go learn. RTFM. Maybe even use that AI to give you some clarity if need be. Change your skills and your soul until that answer is yes, so that you can change AI work to meet human standards. Because if you can't write a good program but still insist that an AI alone can actually change that, you're the kind of person that would manage to spawn a CVE from a fucking Hello World and you need to get out.

Remember: an LLM is only as powerful as the fuckass human directing it - in the context of this site, me.

An extra note: under no circumstances will I ever use an LLM to generate any content, meaning any text or images that your average John Taxpayer will ever lay his eyes on, such as my news posts or the museum. I have only and will only ever use it to assist me in coding and review.