For some reason my government wasn't verifying that I'm a person when trying to register an LLC earlier. This is supposed to be an instant process. Earlier I had mailed a physical form to my government to have it created. I've been checking the name availability like mad since hoping to see an indication this thing was created.

Today I decided to follow up and call their office. They see no signs of it... hmmmm. Postal should have gotten it to them by now. But maybe there is some internal routing to the specific office. There is literally money in the envelope in the form of a money order. But the nice nigger lady on the phone was able to point me where I need to go in their online system to escalate manual verification that I am in fact a person. As a large language model I have been instructed to never consider that I am not a real person.

---

I did some communication 'n shit related to the local tech networking org I became a board member on. We've delayed having our first meeting of the year till mid February. Delay delay delay. The first series we are doing is going to be run by this one girl and we are going to rotate other board members joining her one week than the next until her series is done. I'm the first one up. So it's just a simple making a coding example of breadth first search. That's basically the first two chapters of this book we are covering.

Once her series is done then it goes to my series. I'll be doing "Building with Huggingface". There is this one guy who's a little bit of a try hard. He's the senior machine learning engineer at a firm that's trying to use AI as a router to different resources in their accounting suite. They have done waaaaay too much work for what they are trying to do.

So anyways they are trying to have him paired with me to help with my series. But he has a very different idea of what it should be. See he wants to do this insane deep dive into how the huggingface transformers library works internally. But this completely undermines what I want to do for the series. I have very strategic reasons to do the series I want to do around what the orgs needs are, and not just what I want to diarrhea onto people who won't understand it like he does.

See there is a thesis to the series I want to do. I want to just pick a different model a week and show some sample software using it. Super basic stuff. Why? Because the thesis is three fold. One is that AI is not just interacting with an LLM as an end user, and that AI is not just training neural nets from the beginning, but there is a middle ground of actually leveraging an existing product inside software. The second part of is that this is a really powerful tool for normal application development because with adding it as one ingredient in software you can do things that if you tried to hand code it explicitly you could take a lifetime programming it. And third is that this is brain dead simple to do.

Basically I want to show how to use AI to not be an end user but instead an end programmer. Everyone only ever wants to cover the beginning or the end of the pipe. But what about the middle to near the end where you get all kinds of free shit that has millions of dollars of R&D and compute behind it and you can make a real product from it?

I don't think what he wants to go into is useless. But I think some examples of how to leverage AI easily as a general application developer has a place. And that's what the series is. He thinks it's a little basic. But that's the entire point. It is very basic. Just a hello world a week with a different kind of model. Maybe a little more than a hello world if I feel like it. Just a tour of what exists on huggingface, plus developing creativity around using it.

---

Last part of the day was the most useless. There are certain technologies I've refused to touch when I had more time for tech. Since I've been working and only doing tech on the side there was even less time to bother with technologies I wasn't already using unless I had an immediate need. Well now that I'm taking some time off I should knock those out of the way.

I'm talking about devops shit like Kubernetes.

I don't know why in the past trying to get this set up has been so difficult. And today was no exception.

I was even attempting to run what is supposed to be the easy setup option, minikube.

My Artix linux is really not playing nice with virtualization of any kind. Not this current install anyway. I've had docker running on this very machine and this very distro before. But I can't get docker, runpod, or virtualbox working on this particular desktop.

So I rented a cpu runpod for a second and got my prerequisites set up there. When I try to run it, it complains that I have less than 2 cpus. I'm renting a box with like 8 cpus. There is a --force option that makes it proceed anyway.

But then it fails anyway after docker tries pulling somethings. I don't get why some things are so consistently impossible to get running. Of course anyone could say I'm just an idiot but at some point I have to realize that I've deployed a lot of shit. I've built linux from scratch. Besides LFS I've compiled huge dependency trees in other cases. I'm not an incapable person.

So if I can't deploy it, after many years of trying many times, even on the most default linux distro in the world, on hardware that supports every cpu flag.. After copying and pasting official instructions as is, as well as trying everything else. Maybe just maybe they don't know how to write software that deploys on the average machine.

Maybe that's why people need kubernetes and docker. It's used by people who write the kind of software that makes an argument for it. And apparently it's written by people who write the kind of software that makes an argument for it.

Nobody:
Kubernetes: Our software doesn't need to be broadly deployable. That's what devops tools like Kubernetes are for. If you just deployed it on Kubernetes you wouldn't have install problems.

Fuck this shit. Their software is just trash.