Author	: Bill Kochman
Publish	: Sep. 29, 2026
Update	: Sep. 29, 2026
Parts	: 03

Synopsis:



Dear friends, the three-part series you are about to read is rather unique. That is because while the other 460+ articles and series which you can now find on the Bill's Bible Basics website were entirely researched and completely written by me, other than these introductory paragraphs, as well as the closing paragraphs -- and some occasional editing done here and there by me -- the bulk of this entire series -- I would say around 80% -- was written verbatim by AI, or Artificial Intelligence. In fact, it was collectively written by a total of FOUR different, very well-known AI chatbots. So allow me to briefly explain to you exactly how this series came into existence.

I had approached OpenAI's ChatGPT with a question concerning how weights are used within LLMs, or Large Language Models. As ChatGPT was explaining this concept to me, an interesting idea occurred to me. The idea was inspired by a process that is used within the world of Artificial Intelligence that is referred to as Recursive Self-Improvement. As I explain in my recent five-part series entitled "Conversations With Four AI Chatbots", Recursive Self-Improvement -- or simply known as RSI -- is a repetitive process whereby each new AI model helps to build and improve its next iteration -- or new model -- eventually arriving at a point where human intervention in the building/growing process is hardly required at all.

In other words, one AI model builds the next model. Then that new model will build the next new model after it as well, and on and on it goes. Through this repetitive RSI process, each new AI model becomes both smarter and more capable than its previous version. So, I decided to apply this same concept by assigning a new project to ChatGPT. Rather than just explain weights to me, I asked it to write a fuller explanation. But I instructed it to write it in plain, simple English so that any non-technical, lay reader who knows very little or even nothing about Artificial Intelligence could understand it.

My inspiration for this ChatGPT project actually came from a series of popular books, the first of which was called "DOS for Dummies". Written in 1991 by author Dan Gookin, and then published by IDG Books, the book -- and many copycats which came after it -- became very popular, and eventually turned into a franchise. The basic idea is to write about a certain subject using simple, plain English, without all of the more complex technical jargon which might normally be associated with that subject, so that anyone could understand it.

So, that is exactly what I asked ChatGPT to do as well, but using the topic of Artificial Intelligence and AI chatbots. However, that was just the beginning of my project. You see, after ChatGPT had written his piece, I next gave what he had written to Anthropic's Claude. I gave Claude the following specific instructions to follow:

----- Begin Quote -----

"Pretend that you are a lay reader who knows very little -- or even absolutely nothing -- regarding the internal workings of AI.

Pretend that you are not even familiar with any of the specific words or terminology which is regularly used within the field of AI.

Then, from that perspective of being a complete novice in the field, please very carefully read the text I will share with you below.

After reading the provided text, please further explain any words or phrases which are found in the text which may still not be very clear to a novice.

Likewise, expand on any concepts which may not be completely clear to a lay reader.

You can add additional words, sentences and definitions, as well as provide examples, append new sections to this text, etc. The key point to remember is to make each and every word and phrase simple and easy to understand for a novice. Ask yourself, "If I was not an AI, and if I did not know anything about Artificial Intelligence, Large Language Models, etc., would I understand this word, phrase, method or process?'

So using the above-described perspective, please create a completely brand new, clean version of the following text IN PLAIN TEXT with a copy/paste button. You do NOT necessarily need to remove any of the text which is already there. In fact, I prefer that you don't remove any of it. Simply make the text below better and easier for a non-technical person to understand. Thank you!"

----- End Quote -----

One important point is that I did not tell Claude that I had not personally written the text I was giving him; so he more than likely assumed that I had written it, being as I have shared a lot of my articles with him on previous occasions. My reasoning for withholding this piece of information from him was simple. I did not want his work being influenced by the knowledge that ChatGPT had created the original text, which he was now going to strive to improve.

So Claude's version of the text became the second step in my project to use Recursive Self-Improvement as a means to make ChatGPT's original text even better. But it didn't end there. After Claude had created his improved version of the text, I next gave Claude's improved text to xAI's Grok chatbot. Once again, I gave Grok the very same set of instructions which I had given to Claude. As with Claude, I did not inform Grok that the text had already been worked on by both ChatGPT and Claude.

And thus, the third step in my project to utilize Recursive Self-Improvement to create a "Dummies" AI guide for novices was completed. Once Grok had completed his improved version of the text, which itself was based on the improved version which had been created by Claude AI, which was based on the original text which had been written by ChatGPT, I next gave the twice-improved text to Google's Gemini. Once again, he was given the very same set of instructions as I had already given to ChatGPT, Claude and Grok. Furthermore, Gemini was not told that ChatGPT, Claude and Grok had already worked on the text which I had before him.

Lastly, once Gemini had created his improved version of the text -- the fourth step in the improvement project -- which became the newest iteration of all of the previous versions of the text, I gave his version back to ChatGPT for a final text improvement cycle. ChatGPT had absolutely no idea that Claude, Grok and Gemini had worked on the same text as well. And thus, I ended up with the Artificial Intelligence guide which you will now read below. As I already said, about 90% of it was the direct creation of ChatGPT, Claude, Grok and Gemini, with some additions, editing and formatting by me.

My friends, I hope that you find this Artificial Intelligence manual both informative and educational. It is really quite interesting to say the least. I think that it will truly blow your mind once you better understand what is happening under the hood whenever you are chatting with an AI chatbot. As I explain in "Conversations With Four AI Chatbots", as well as in "AI, Chatbots, Daemons and Demons", it has nothing to do with spiritual forces or being possessed by demons or talking to demons. It is all about pure science and some very complex mathemtatical operations. Not just a few of them, my friends, but literally millions of them. Knowing this makes it easier to understand why LLMs require so much computing power -- or compute -- and electrical power in order to even operate.

----- Begin Quote -----

HOW CHATBOTS UNDERSTAND AND WRITE LANGUAGE

-- What is Under the Hood: A step-by-step, plain-English explanation for anyone who has never studied computers or delved into the amazing world of Artificial Intelligence

INTRODUCTION: WHAT IS AN AI CHATBOT?

An AI chatbot is a computer program that can communicate with people using ordinary human language. It relies on what is referred to as a "Large Language Model". But what exactly is a Large Language Model?

A Large Language Model -- commonly abbreviated as an "LLM" -- is a type of Artificial Intelligence system that has been trained on enormous amounts of written language so that it can recognize patterns in language and generate new text. The word "large" refers primarily to the enormous size and complexity of the mathematical model, including the huge number of learned numerical settings it contains. The words "language model" refer to a system designed to work with language by calculating which pieces of language are likely to follow other pieces of language.

In simple terms, an LLM is the powerful AI system that does much of the language processing and generation behind an AI chatbot.

Think of it this way: the chatbot is the part you interact with, while the LLM is the underlying AI model that actually processes your words and generates much of the response you receive. The chatbot provides the conversational interface -- the place where you type your message or prompt, and then receive an answer -- while the LLM provides the underlying language-generation capability.

This is somewhat like the difference between a car and its engine. The car is the complete thing you interact with and drive, while the engine is a major component inside it that provides the power that makes the car move. Similarly, an AI chatbot is a complete interactive application, while an LLM can be the underlying AI model that provides its ability to process and generate language.

It is also important to understand that an AI chatbot can contain more than just an LLM. The complete chatbot system may include additional software and tools that handle things such as the conversation interface, instructions, memory or conversation history, safety systems, access to external information, and other functions. The LLM is the central language-processing component, but it is not necessarily the entire chatbot.

Examples of chatbots include ChatGPT, Claude, Gemini, Grok, Meta AI and other similar AI systems. You type a question, request, or statement into the computer, and the AI produces a response in words.

To a person, this can feel surprisingly similar to having a normal conversation with another human being. However, what is happening inside the computer is very different from what happens inside a human brain. An AI language model does not read a sentence in exactly the same way that you do. It does not possess human thoughts, human experiences, or even human understanding of the world. Instead, it processes language by performing enormous numbers of mathematical calculations.

That may sound intimidating, but the basic idea can be explained fairly simply. The AI takes the text you give it, breaks that text into smaller pieces, converts those pieces into numbers, processes those numbers through a huge mathematical system, and then calculates what piece of text should come next. It then repeats that process again and again until it has produced a complete response. This guide explains that process step by step. Along the way, we will encounter some technical terms that are commonly used when discussing AI, including:

• tokens

• token IDs

• embeddings

• neural networks

• weights

• parameters

• training

• context

• attention

• transformers

• probability

• prediction

• inference

• generalization

• hallucination

These words can sound complicated because they come from the technical world of computers and mathematics. But the basic ideas behind them are much easier to understand than the terminology might suggest. So let's start at the beginning.

STEP 1: BREAKING YOUR MESSAGE INTO "TOKENS"

-- Turning Text Into Small Chunks

When you read a sentence, you naturally recognize words and their meanings.

For example, if you see . . .

"The cat is sleeping."

. . . you don't have to consciously break the sentence into pieces. You simply read it and understand it.

A computer works differently.

TOKENS:

Before an AI language model can process your message, the text must first be broken into smaller pieces called TOKENS.

A token is simply a small piece of text.

Depending on the particular AI system and its vocabulary, a token might be:

• an entire word: cat

• part of a word: play + ing

• punctuation: ! or ?

• a number

• a space or other small piece of text

A token is therefore NOT necessarily the same thing as a word.

Sometimes one word is represented by one token.

Sometimes a long or unusual word is divided into several tokens.

For example, the word "playing" might be represented by pieces roughly corresponding to "play" and "ing."

The exact way a particular word is divided depends on the AI model's tokenizer, which is the computer system that performs this chopping-up process.

ANALOGY: LEGO BLOCKS

Imagine that you are building a house out of LEGO blocks.

Some LEGO pieces are large.

Other pieces are small.

You can combine those pieces in different ways to build something much larger.

Tokens work somewhat like those LEGO pieces.

The computer takes your message and breaks it into manageable pieces.

Those pieces can then be processed mathematically and eventually put together again to produce a response.

For example:

Your sentence . . .

"The cat is sleeping."

. . . might be broken into several tokens representing pieces of:

"The"

"cat"

"is"

"sleeping"

"."

The important point is that the AI does not directly manipulate your sentence as a human reader does.

It first converts the text into smaller pieces that its mathematical system can work with.

TOKEN IDs: GIVING EACH TOKEN A NUMBER

Once the text has been divided into tokens, the computer needs a way to identify each one.

So each token in the model's vocabulary is associated with a number called a TOKEN ID.

For example, imagine that a particular AI model has assigned:

"cat" = Token ID 4128

"dog" = Token ID 5901

These numbers are only examples. The actual numbers used by real AI models are different.

A Token ID is basically a label.

Think about checking your coat at a theater.

You hand your coat to the attendant, and you receive a numbered ticket.

Suppose your ticket says:

42

The number 42 does not tell you that your coat is black.

It does not tell you whether the coat is made of wool.

It does not tell you whether it is warm or waterproof.

It is simply a label that allows the theater to identify your coat.

A Token ID works in much the same way.

The number itself does not contain the meaning of the word.

It simply tells the computer which token it is dealing with.

This distinction is important:

Token = a piece of text

Token ID = a number that identifies that piece of text

STEP 2: "EMBEDDINGS"

-- Giving Tokens A Mathematical Location

Now we have a problem.

The Token ID tells the computer WHICH token it is dealing with, but the number itself doesn't tell the AI very much about the token's relationships with other words.

For example:

"cat" might have Token ID 4128.

"dog" might have Token ID 5901.

But the numbers 4128 and 5901 do not inherently tell the computer that cats and dogs are both animals.

So the AI performs another conversion.

Each token is represented by a long list of numerical values called an EMBEDDING. In other words, an embedding is actually a mathematical representation or description of a token.

These are ordinary mathematical numbers -- often positive and negative decimal values -- not simply strings of 0s and 1s. Although computers ultimately store and process all digital information using binary at the hardware level, the AI's mathematical calculations are normally described using these higher-level numerical values.

The individual numbers do not normally correspond to simple, human-readable attributes, such as "animal," "furry," or "four-legged." Instead, the entire collection of numbers works together to represent different characteristics and relationships which are associated with that token.

So, thus far we have the following:

Token = a piece of text

Token ID = a number that identifies the token.

Embedding = a mathematical representation of the token.

Binary = the underlying digital representation used by the computer hardware.

A useful way to imagine an embedding is as a location on a gigantic mathematical map.

This isn't a physical map that you could unfold on a table.

It is a mathematical "map" containing many dimensions that humans cannot easily visualize.

Words and pieces of text that are frequently related to one another tend to develop mathematical representations that are related to one another.

For example:

CAT and KITTEN are closely related.

CAT and DOG are also related.

CAT and SKYSCRAPER are generally much less closely related.

The important thing is not that the computer has been given a simple dictionary definition saying:

"Cat = small furry animal."

Instead, during training, the model develops mathematical relationships between words and pieces of language based on the enormous amount of text it processes.

The embedding is one of the ways those relationships are represented mathematically.

ANALOGY: A MAP

Imagine a gigantic map.

On this imaginary map, places representing related concepts tend to be located relatively near one another.

The area containing "cat" might be close to areas associated with:

kitten

dog

pet

animal

fur

whiskers

Other concepts might be much farther away.

Again, this is only an analogy.

Real AI embeddings involve very large numbers of mathematical dimensions and are much more complicated than an ordinary two-dimensional map.

The important idea is simply this:

The AI converts pieces of language into numbers that allow the computer to mathematically compare and relate those pieces.

So far, the process looks roughly like this:

YOUR TEXT

↓

BROKEN INTO TOKENS

↓

TOKENS GIVEN TOKEN IDs

↓

TOKENS CONVERTED INTO EMBEDDINGS

↓

NUMERICAL REPRESENTATIONS READY FOR FURTHER PROCESSING

Please go to part two for the continuation of this series.

⇒ Go To The Next Part . . .

---

---

Please avail yourself of other areas of the Bill's Bible Basics website. There are many treasures for you to discover.