Introduction — Getting the Big Picture Before You Use AI

Using AIAI is a broad field/concept covering technologies and mechanisms used to tackle intellectual tasks with computers and the like. Prediction, classification, and generation are examples of this, and machine learning is one representative method related to AI. AI as a whole is not the same thing as machine learning, generative AI, or LLM. to write text, generate code, or help with research and development has stopped being anything special. So do the AIs we use every day really "think" and "make decisions"? Why do they sometimes give plausible but incorrect answers, or say that they have "checked" something when they have not actually verified it?

Are "AI" and "LLMA Large Language Model is a language model treated as large-scale in current technical usage, used for a broad range of language-related tasks. There's no fixed numerical threshold — by parameter count or data volume — that universally defines "large" here. LLM is not the same thing as AI as a whole, a Chat Model, a Foundation Model, or Transformer itself. → Language Model" the same thing? When an AI says "I searched for it," can you actually confirm that a search took place? When it says "I remember," what exactly is stored, and where? When it reports "I ran the tests. All PASS," can you treat that sentence alone as proof the work is done?

This book covers the basics you need to answer these questions.

The purpose of this book

This book is not a specialized textbook for studying AI itself. The goal is to help you organize, on your own, the following points when using AI in AI-assisted development:

• what the AI is actually doing
• what can be confirmed from the outside
• what remains inference
• what is still unknown
• what a human ultimately needs to decide

We start from the basics of AI and LLMs, then move through how a model generates output, TrainingTraining is the process of adjusting a model's parameters or learned state, based on data and a learning objective. Training is distinguished from Inference, which uses a trained model to obtain predictions or output. Training as a whole is not the same thing as just Fine-tuning. → Parameter, Inference, Fine-tuning and InferenceInference is using a trained model, by giving it input, to obtain predictions or output. Its role differs from Training, which adjusts a model's parameters, and it isn't a term that refers only to text generation. → Training, Input, Output, ContextContext is the information available to a model or system for the current processing or inference. It may include things like the current input, prompt, or conversation history, but what's included differs by system. Context is a separate concept from the context window (the upper limit on how much can be processed), memory (retained for later reuse), and training data itself. → Context Window, Prompt, Memory and Memory, SearchSearch is a process/feature that takes a query against a collection of information or documents and finds information or results relevant to the goal. Web search is one example of Search, but Search does not always target only the web, and it is not the same thing as Generation or Retrieval. → Retrieval, Generation and Tools, EvidenceEvidence is the grounds or observation results used to confirm a Claim. In this book, we consider how far something can be confirmed, in light of the object, the method, the scope, and the conditions. AI-generated text, strongly confident wording, agreement among multiple AIs, or a generated citation, by themselves, do not become Evidence that independently confirms an external fact., Source of TruthA Source of Truth is an information source chosen as the standard for confirmation, for a given question or Claim. Depending on what you want to know, the appropriate source of confirmation varies — a specification, a file in the repository, Git state, a test result, a CI result, a configuration, a runtime observation, a log, and so on. There is no single, universal Source of Truth, and an AI does not get to decide its own confirmation source and gain authority over it., and Verification. By the end, the goal is not just to know AI terminology, but to be able to separate what an AI's generated explanation says from what has actually been confirmed.

Understanding AI and confirming AI

An AI having "said" something and that content actually being correct are not the same thing. For example, suppose the following sentence is generated:

I ran the tests. All of them returned PASS.

That sentence alone does not let you confirm which tests were run, against what target, or under what conditions. At the same time, it is also not appropriate to treat everything that cannot be confirmed from the outside as simply wrong.

So this book distinguishes between:

• what has been confirmed or observed
• what has been inferred from evidence
• what has not yet been confirmed

Later chapters organize this into the classifications VERIFIED / OBSERVED, INFERENCE, and UNKNOWN / UNDISCLOSED. You don't need to memorize the terms yet. For now, just carry this feeling as you read:

"What the AI answered" and "what could be confirmed externally" may be two different things.

Who this book is for

This book is aimed primarily at:

• people just starting to use AI-assisted development
• people relearning AI or IT from the basics
• people who want to use AI in development, infrastructure, or operations

It does not assume you have read specialized textbooks on AI or machine learning. It is designed so you can read through without needing Transformer mathematics, neural-network math, hands-on experience with Agents or RAGRAG is a method/configuration that obtains relevant information from an external source and uses that information to perform generation. It's an approach that combines Retrieval and Generation, but using RAG does not guarantee the factuality of the output, and it is not the same thing as retraining the model. → Retrieval, Generation, Grounding, Factuality, MLOps, or vendor-specific certification knowledge.

Making the book self-contained

When you're learning about AI, unfamiliar terms keep appearing one after another. A textbook that forces you to search external sites every time you hit one is a burden for a beginner. For Version 1, this book aims to let you read through the basic learning goals using the book alone as much as possible. External materials and primary sources are important, but they are treated separately from the main text — as grounds for verification and further study.

If you get lost on a term or a distinction in the main text, you can consult the Appendix at the back. Appendix A is a glossary for looking up basic AI terminology; Appendix B is a definitions reference for confirming the meaning of this book's own terms, such as verification, Evidence, Source of Truth, and Human authority. Note that Chapters 13 and 14 use basic software-development terms such as repository, commit, push, test, and deploy. This book does not teach version control itself — its focus is on how those states relate to working with AI.

How to read this book's diagrams, categories, and internal structures

The conceptual diagrams in this book are meant to organize ideas and relationships — they do not represent the actual internal structure or processing order of any specific service. Likewise, the categories used for learning are not necessarily official industry-wide classifications. Where no public information or observable Evidence exists for an internal mechanism, we do not fill the gap with speculation; we treat it as UNKNOWN / UNDISCLOSED.

What this book does not cover

For Version 1, the following are not goals:

• explaining the mathematics of Transformers or neural networks in detail
• writing a complete history of AI
• speculating about the undisclosed internal structure of specific AI services
• writing an operations manual for a specific product
• assuming that every AI service works with the same internal structure
• filling in unconfirmed matters about AI with plausible-sounding explanations

Where a mechanism is not publicly disclosed, we leave it as UNKNOWN / UNDISCLOSED as needed.

How this book proceeds

First, we build a map for thinking about AI. Chapter 1 separates easily confused terms such as AI, Machine Learning, Model, LLM, and "service." Chapter 2 looks at the history leading to today's AI not as a simple "old technology replaced by new technology" story, but as a map of multiple ideas that developed while overlapping one another. Chapter 3 looks at AI services from the user's side and organizes what can be observed and what remains unconfirmable.

After that, we move through model-based generation, Training and Inference, work that is easy to delegate to AI versus work that needs external confirmation, Context and Memory, Search / RetrievalRetrieval is finding and taking out documents or information relevant to an information need, from a collection or index. It is closely related to search, but it is not Generation itself, and Retrieval alone does not mean RAG as a whole. → Search, RAG, Generation / Tools / Agents, and the physical environments AI runs in. In the latter half, we cover the problems that remain in current AI, Evidence, Source of Truth, and how to use AI in AI-assisted development, before finally following one AI-assisted development task from start to finish. You don't need to understand everything from the start — the distinctions built in earlier chapters get used gradually in later ones.

This book also checks its own evidence as it's built

This book was not written by drafting the main text first and then hunting for convenient grounds afterward. Roughly, it was produced in the following order:

decide what should be taught
        ↓
decide the questions that need investigating
        ↓
gather Evidence
        ↓
decide Claims within the range the Evidence supports
        ↓
write the beginner-facing text
        ↓
review the content and expression
        ↓
Human makes the final decision

You don't need to remember this production process itself. What matters is that this book's own explanations are also not treated as "correct because they sound plausible."

First, start by separating the words

When thinking about AI, the easiest place to stumble is treating words that refer to different things as if they were the same. Chapter 1 begins by organizing exactly what the word "AI" refers to.