Modern Talking; Or, How AI Is Helping Me Jump Ahead Of Myself

The Beowulf computer cluster has been on my todo, for multiple decades now.

I assumed, I’d never have time to get to it.

Recently I begun work on Queue Processing, in my Visual Programming language, the thing that makes me busy.

And with no more than 30 seconds to myself, I quickly prompted AI for a Beowulf cluster in JavaScript.

I chose JavaScript, because it is a simplified C family language.

AI aided languages with AI based memory management, are sure to be even simpler come future.

And just like that the AI gave me an HTTP master server, and workers, deciding that the master will distribute tasks.

I asked about master re-election, in an unreliable network. and it clearly listed re-election strategies.

In under a minute, I took care of a really old TODO, that I would never have time for.

The result of having the opportunity to look into Beowulf Clusters, and being shown how they are just a few lines of code.

Is me looking for integration strategies, in my visual programming language.

I already have local and network queue processing, but I want the backing of an older approach.

I want to have a Beowulf Cluster project in my templates, and an article with multiple strategies.

None of this would be possible without an AI assistant, who speaks in my language of choice.

When I ask for JavaScript examples of common problems, I get 99.9 accurate results.

But when I stray asking for code that usually lives, in other languages, there maybe a line missing, but it is easy to fix.

The Beowult cluster question, strained the AI, as it is an idea, not a program.

But with some encouragement I got a 75% working sample, that would need less than an hour of to to get running.

When I add “in JavaScript” I expect AI to “translate”, which is what LLMs are good at.

I previously tested translation of computer programs, into declarative XML with good success.

And had excellent results with LLMs frigteningly understanding, not only translating into an EventEmitter, but an Agent.

Event Emitters have an on and emit action, so you can say on hello say meow.

And then just state, emit hello, and the agent will output meow.

Very clear, simple, and powerful.

An agent has an on input, and on upgrade, where data emitted into the input...

Is processed by code, sent in via upgrade.

Thus, when a program made of agents is running, you can upgrade its parts, its agents.

In real life an agent already has code, and upgrade is something that is used to just upgrade it.

But that is only done, not to scare programmers, and is not needed.

Simple and clear is good for coders and AI, I don’t think AI follows the flow as I explained it.

It is likely paying attention to the words input and process, but as AI get faster it will.

The clearer the code, and the simpler the language, the finer the AI contributions in the future.

Finally, simplicity and clarity throughout adds up, it helps you read code faster, it helps AI be more accurate.

It helps you with back of envelope flow control and architectures, and it helps AI to dream out clear starter code.

For me specifically, as I tune and refine my Visual Programming Language,

JavaScript is already too complex, any code I write in JavaScript, in the visual programming world real I’ve built…

Is either unnecessary, or will be replaced.

In fact, I already made the wrong assumption, that I must replace JavaScript with Gherkin.

Which can be thought of as structured English, or a subset of English.

Nope, the units of visual programming, must be a plain English description of code.

So that AI not only generates it, but re-generates it as it gets better.

Even free AI is perfectly capable of generating a list, of commony used visual programming blocks, of functional programming funcions.

And based on that list it will generate EventEmitter based Agents, with proper tests, and even a README file.

Not only my tests show almost no code hallucinations, but in one test the AI dropped into reusing other libraries in my blocks.

Effectivley wrapping Ramda-js and lodash functions, with an agent making all these functions part of an EventEmitter network.

In a ward, they can be used raw, to create drag and drop programs.

Visual programming it self, multiples all this power, the lines that the user connects boxes with…

Is all that is needed to create programs, that go far beyond extracting and transforming data.

Interacting with a tree of web components in a web page, creates a user interface development environment.

And again, one that is clear to understand, and is likely not to pose a challenge to AI.

The simplet the language, the more effective the AI.