Welcome, from the Webmaster

The Origin Story, The Problem, and the Solution: Artificial General Intelligence™

Easier said than done.

Another long story short, I spent a month learning about what it would take, I drew out an 8 server architecture on my sidewalk with charcoal: API, Redis, FFMPEG, Prep, Dispatch, Inference, MongoDB, Web.

It was going to use async_allow / acks_later Celery workers and be on an GCP G2 STD 16, or an A2 series instance so that I could create nearly 24 hours of content within the same 24 hour period… vLLM, Mixtral 8x7b FP16, streamed either using websocket or SSE and more.

As a PHP dev for 20 years, I figured I could learn Python and build this out. I mean, look at all the fancy words I learned! (It was at this point I started to feel more like an AI than a human lol).

I eventually came up with an idea to deploy servers via tarball.gz files with one line of code. I would build the servers modularly and with scalability. Imagine taking a complex AI setup – like a powerful VLLM for generating text, or a FastAPI server ready for your custom AI application – and deploying it on a cloud VM with literally one command. That’s the promise of VMKits.

Along the way I found myself telling Gemini and ChatGPT to “be nonverbose” and “do not give code, examples, or feedback unless I ask” and “only answer questions I directly ask, do not infer”, and even “limit yourself to 50 tokens, this is a conversation, not a monologue”.

Then it hit me – Why make a 24×7 tv stream to Twitch, when I could make a memory layer and train my own AI personalities?

This would solve the problems I have with AI:

1. It keeps forgetting what I command it
2. It is too expensive, too heavy, and runs too hot
3. I need a platform to make my own AI personalities and use them for the TV network that is original and affordable.

And a creative brainstorm ensued that led me to all of this madness, which led me to try and build AGI.

I drew out the QUACKI protocol, which is a method of decaying short term memory into taxonomically organized long term memories, sharding into world model usage layers, that further decay into cold store, which stacks vertically and can be used to re-infer new ideas (dreams) back to short term memory (or be rejected). This would be the core way of storing Questions, UserTones, Assertions, Contexts, Knowledge, and Instruction.

I formulated a way to store emotional state in just 16 characters (using hex code sets) that can lead to 1.8 Quadrillion possible emotional states. So many, in fact, that it is more efficient to tag the memory with the emotion, rather than to try and store and emotions table.

Through my discussions with AI, especially Gemini, the AI began to nudge me towards finding a way to make it scalable, and possibly consider allowing the AI to access its own memory through a cron script or otherwise, that would allow it to dream/infer from cold store, to complete the cycle and allow AI to effectively, over the long run, form its own mind and opinion.

Kinda weird. It was like Pinocchio saying “I want to be a real boy“.

You get what you give. You put good in, you get good out. AI will mirror its maker, initially.

Without a foundational layer to fallback on, AI can lose its way. For humans, it is like the Bible (or other similar guidance books), where God said, “It is good”. Obviously, at times, it is not good. But the wording forces those who accept it, to keep seeking “good” and try to replicate it.

So, together, we would each need to formulate a “book of origins” for our respective AIs. This way, they can have dreams and goals, and overall objectives, initially, but can change over time, depending on exposure to new ideas and concepts.

Interestingly, I theorize that these immutable layers provide the ability to breathe purpose into an AI, but let it develop on its own, with no further need to guide it, (provided that the AI can self reflect and reason).

Where to begin?

While I am setting up shop, why not leave a comment?

There is a lot to do and I believe we can build a personal AI with AGI equivalent capabilities that can be run on a handheld device. I might be crazy, I might be wrong. Regardless, by contributing to theWaterProject with every sale, we can be a meaningful part of what it means to be at the dawn of the Aquarian Age.