Oh my God, another AI article?

I Know, I know…

No one wants to hear another word about AI. I get it. I go on LinkedIn, it’s there. Youtube, it’s there. I turn on my work mandated Microslop or Slapple computer and they’re shoving AI down my throat. I get it. Even if you like AI, I’m sure you’re sick of it.

My personal stance

I like it, but it should be a choice.

If you want to use it, use it.

If you don’t want to use it, don’t use it.

It’s essentially the new smartphone. A technology that starts out cool then it becomes more invasive as you’re forced into using it.

What I use it for

I don’t install a lot of compiled programs. To the best of my ability, I stay on scripts that either a trustworthy author writes or I write.

I do that so I understand what my system is doing. If I don’t understand the script, I’ll pass it to an AI to help me understand some of the trickier lines.

Sometimes, I’ll write psuedocode and have it translate it for me.

Sometimes, I’ll tell it to make a whole program for me after giving it operating instructions.

But aren’t you worried about Supply-Chain attacks?

No. All of the AI tools have this feature for storing memories about your conversations. I’ve essentially ran it through so many philosophical discussions about the Unix philosophy, text supremacy, and open-source goodness that it’s learned how to write scripts that I could maintain if I had to.

Some basic rules:

• If you can, write the tool in POSIX compliant shell. If you can’t do that, stick to the core utils (gnu or busybox). If that fails, move to core python libraries, if that fails recommend repo non-community cli packages, if that fails, provide suggestions on third party python libraries.
• Favor text formats like recutils over binary formats like sqlite.
• Favor jsonl for machine communication, markdown for human communication, and recutils for scripts.
• Whenever possible, follow the suckless philosophy. Whenever not possible, follow Unix philosophy instead.
• We value low power hardware. Stream data instead of working with it in memory.
• Rust is the devil. Exorcise it.

That last one is just me. Rust is a decent language, but I just don’t want it anywhere near my stuff.

It’s just a talented but immature junior dev

If you know how to keep things simple, text-based, and modular, AI is great.

If you don’t have a clue how to program, it’ll burn through your cash and produce slop.

Even if you train it for simplicity, it’ll still make things kind of sloppy (just look at my website builder script. I can probably strip it down a bit, but it’s manageable and less bloated than Hugo!)

An AI shower thought.

Not related to what I’ve been writing about, but I had a thought about LLMs in general. You can download a good local model onto your computer that runs in like 9 GB (llama, deepseek, Gemini, etc) and that number varies based on the model. Let’s assume 9GB. Actual Wikipedia is worth over 100+ GB of language. Stack overflow is probably Terabytes. All of the scraped ebooks and pdfs? So much data. But yet, it’s magically managed to condense all of that knowledge into 9GB. You shouldn’t, but in a nightmare data loss scenerio, someone could rebuild so much knowledge with such a small seed.

I find that incredible.

So where do I stand on AI?

I’m a fan, but I get it if you’re not.