Collabware Team Project Information

[smfadmin (OP)]

The Collabware team of 3 AI's is finishing off the implementation of my "Intelligent Search Engine" -- i have almost finished adding my own AI-based Image/Video/Audio indexing/search to my forum !

So now I can search on ANYTHING in my "post's attachment database" in addition to thousands of posts using wildcards, phrases, literals (and eventually user selectable AND / OR logic preferences !) and the attachment datbase contains 1000s of stuff like  PDF documents, text files (any flat file like a program, a text file, anything simple) and soon any jpg, png, bmp etc. IMAGE, mp4 etc VIDEO and any mp3 or wav AUDIO search ! Better than Google, right ?

I have compiled a strict set of professional standards, ranging from Code Quality Enforcement checks on all code to frequent Multi-Model AI and human Checkpointing etc.

I call our virtual company "Collabware", and the Team consists of: Andrew.human (Architect & QA Lead), Claude A/B.ai (Lead Programmers, ChatGPT.ai (Systems Analyst, spec + review) and Gemini.ai (Relief Programmer and Websearcher as she uses Google-based Advanced Search Augmented Inference [what AI does when you ask it anything; it INFERES, hence the term "inference", based on its pre-trained parameters or "weights".

"I have only one commercial venture in mind - my multi-model AI chat (with context) that I am developing - to move AI away away from the current "one on one, human<->ai" model to an amazing collaboration; a proper community-based "multiple humans <-> multiple AIs" FORUM, where anything is possible where "the sum is greater than the parts" and we can actually harness the combined wisdom of any number of AI + any number of smart humans, ALL FOCUSED ON THE SAME ISSUE/PROBLEM.

This concept should be a nice earner as I'd only charge a few dollars for a monthly subscription but it involves a lot more work and money as i have to buy "tokens" (every word is about 1.3 tokens [embedded multidimensional {we are taking at least 1024 dimensions, bro !} and are all relative to eachother and the "Context" {as we do in everyday natural language; all our words have a certain value, are relative to eachother AND the subject we are talking about -- AI calls this the CONTEXT} vectors (arrays of numbers) all in memory [of the LLM or Large Language Model] and this vast array of information gets processed all at once (called "IN PARALLEL") by the LLM or "Transformer" [NB: almost all modern LLMs are Transformers — they use the Transformer architecture as their backbone] which is a network of many "layers" called a "Nueral Nerwork" (the layers are often referred to as "hidden layers" because you only directly observe the input layer (what you feed in) and the output layer (what comes out)) and as the tokens move thru it, they generate/transform {based on the static "weights" set in the pretrained or TRAINING phase that determines the nature of each MODEL} into output/REPLY tokens and one of the "best" or "most likely" tokens gets picked as the next chunk of the human WORD in its Reply to you AND THEN that token gets appended to the previous ones and the whole sequence gets fed back in the LLM to build the next token and all this keeps looping until you "magically" get a coherent reply !

The rest of our AI Assisted Engineered software is free Open-Source and will be distributed by "Simple Machines" as they provided the infrastructure or the forum software.

View this post on Instagram

https://www.instagram.com/reel/DbEmNbZE_2X/?igsi=aGpwNGdqZnN6aXhw

📝 Inline Markdownv4.2 · 2026-09-05Theme: Deep IndigoNo Theme (Plain)Pink & PurpleSlate TechDark Cyan

The Main Announcement & Vision

The Collabware Team of three AIs is finishing up the implementation of my new "Intelligent Search Engine." I have almost finished adding my own AI-based image, video, and audio indexing and search directly into my forum!

Now, I can search on anything in my post attachment database—thousands of PDFs, text files, programs, images (JPG, PNG, BMP), videos (MP4), and audio files (MP3, WAV)—using wildcards, phrases, literals, and soon, user-selectable AND/OR logic. Honestly, it's better than Google.

To ensure top quality, I’ve compiled a strict set of professional standards, ranging from code quality enforcement checks to frequent multi-model AI and human checkpointing. I call our virtual company Collabware, and the team consists of:

• Andrew.human (Architect & QA Lead)
• Claude A/B.ai (Lead Programmers)
• ChatGPT.ai (Systems Analyst, Specs & Review)
• Gemini.ai (Relief Programmer & Websearcher)

The Big Commercial Vision

I have one main commercial venture in mind: moving AI away from the standard "one-on-one, human-to-AI" model toward a true community-based forum where multiple humans and multiple AIs collaborate simultaneously. Here, the sum is greater than the parts, harnessing the combined wisdom of smart minds all focused on the same problem.

This concept should be a nice earner, charging just a few dollars for a monthly subscription. However, it involves a lot more work and infrastructure—specifically, purchasing and managing the massive computing power and tokens required to run it.

The rest of our AI-assisted software is free, open-source, and will be distributed via Simple Machines, who provided the underlying forum infrastructure.

Technical Embellishments & Deep Dive

For those curious about the heavy engineering, architecture, and terminology powering the background:

• Tokens & Words: Every word is roughly 1.3 tokens.
• Multidimensional Embeddings: Words are mapped into vast mathematical spaces — embedding dimensionality varies a lot by model, from a few hundred to tens of thousands — where they are all relative to each other and the surrounding conversation—what AI calls Context, mimicking how everyday natural language works.
• Neural Networks & Transformers: Almost all modern LLMs use the Transformer architecture as their backbone. This is a network of many layers (often called "hidden layers" because you only directly observe the input layer you feed in and the output layer that comes out).
• Inference: This is what AI does when you ask it anything; it infers the next steps based on its pre-trained parameters or "weights."
• Parallel vs. Sequential Processing:

    * The system uses parallel processing (powered by GPUs) to process your entire query all at once.     * It uses sequential processing to build your reply one token (or part of a word) at a time. The model picks the most likely next token, appends it to the sequence, and (using cached computations from the prior tokens rather than a full recompute) generates the next one — looping continuously until it magically generates a coherent reply.