What Is It That I Am Personally Doing With Artificial Intelligence?

I am building software that does not mess around, it does not make assumptions about anything working well.

Currently in my laboratory, I have a virtual world, that starts with a lobby, and allows me to enter room A, and then room B.

Room A, has Enter Actions that includes, starting PersonaLive AI video generator which eats half of my video memory.

Room B, eats all of the memory, by starting up the open source GPT, that I am experimenting with locally.

If I went straight from lobby to room A, and room B, my computer would crash with an Out Of Memory (OOM) error.

I designed the two rooms to pose a real danger, and I used an older Artificial Intelligence technology to make things safer.

First of all, I am doing this because I believe local AI can be made smarter, than anything else out there.

Fundamentally, most of us complexity nerds believe AI was a solved problem, but it needed a lot of typing.

Thus we see the large language models are the content creators for the finer, STRIPS related software.

STRIPS stands for Stanford Research Institute Problem Solver, but this is not about Stanford, many people came up with these ideas.

Roughly speaking, you say, I need to have a Shelf for my plant, and we start, with a no shelf state.

Then standard computer programming creates a plan to build a shelf, out of Actions, that long ago needed to be manually typed in.

Actions like hammer the shelf together, would have JUST two things:

a pre-requisite, nails, wood, and a result, shelf.

An action like go to hardware store, would have no pre-requistes, and result in nails and wood.

With simple programming, a planner would make a plan, to go from “no shelf state” to “I have a Shelf for my plant”.

That would expain the actions that must be taken, by following the pre-requisites and results.

And we end up with “go to hardware store” and “hammer the shelf together”.

That is absolutley artificial intelligence, but it needs actions, it needs somone to type up the pre-requisites and results the first time.

And that someone is the large language model, so we had AI, but it would take forever to do all the typing for different things.

Now in some cases, like linux system management, for example, we only need the AI once, at first.

And after that it is all simple, basic programming, probably 50 lines of code.

So I am not going anywhere, without the old AI, because it allows me to do amazing things and only needs the new AI once.

And please note, we never which which hardware store, so actions can work on top of what is available, not specifically required.

In my operating system layer, virtual machine use, where I emulate a computer within a computer, in one way or another.

I tent to use one of three programs, VirtualBox, docker, or podman, this ancient AI can choose the right, or the available technology.

Aside from rettry and few other things, actions have a cost, VirtualBox is cost 3, it is a full GUI OS, podman is cost 1, it is efficent.

When available, this old primitive AI, will look at podman first, then then check for docker, and only as a last resort use virtual box.

At the moment, in my rooms, I am using the notion of state such as started, ratehr than the idea of commands: like start.

Because when you enter a room twice, you don’t want to run the start command twice, that would again OOM.

You want to say, I need state, started, and that is how the old AI operates: states.

So when you enter a room twice, the second time my GOAP or STRIPS server just says state already reached.

It is worth mention here a the end, that the UI for this thing, the entire programming system, can be operated by a child.

It is a set of rooms, connected together by arrows, with enter and exit lists, to which you drop actions into, from a tidy library.

The action for starting a persona server, has just one UI component, the dropdown box where you specify state, started, or stopped.