AI 101 — A Practical Guide for People Who Will Actually Deploy It

In everyday conversation "AI" means three or four very different things. This chapter lays out where each term sits — from classical ML up through LLMs, VLMs and VLAs — and settles the vocabulary people most often confuse, like Dense versus MoE.

Contents

1. 01 How Many Kinds of AI Are There? The Whole Map In everyday conversation "AI" means three or four very different things. This chapter lays out where each term sits — from classical ML up through LLMs, VLMs and VLAs — and settles the vocabulary people most often confuse, like Dense versus MoE.
2. 02 Reading a Model Name Model names are not arbitrary. They are a code that tells you almost everything worth knowing before you download. This chapter decodes one piece at a time, with tables for suffixes, file formats, and the quantization level that decides whether a model fits your machine at all.
3. 03 Hardware — Choosing a GPU and a Machine Most people pick a GPU by its TFLOPS figure, which is the wrong number for inference. VRAM decides what you can run at all; memory bandwidth decides how fast it answers. This chapter separates the three numbers, tiers the hardware by budget, and gives a break-even formula for buying versus renting.
4. 04 What Kinds of Runtime Are There? The stack from driver to user interface, which runtime suits which job, and the real commands for Ollama, llama.cpp, vLLM and SGLang — plus why your application should speak a standard API from day one so you can change runtimes without rewriting anything.
5. 05 Measuring What a Model Actually Consumes The VRAM formula and the KV cache that grows with every concurrent user, the nvidia-smi commands worth knowing, which numbers to read and how to interpret them, and how to benchmark without fooling yourself.
6. 06 Choosing a Model That Fits the Job Start from your constraints, not from a leaderboard. A job-to-model table, the one rule that always holds, how to build your own eval set, what is specific to Thai, and how to read a licence before you use a model commercially.
7. 07 Where Models Come From, and Downloading Them Safely The main model sources and what each one is, how to read a repository before you download it — the good signs and the dangerous ones, one by one — pinning a version by commit sha, and the limit of pinning that you need to understand.
8. 08 Security — Where the Danger Actually Is The risks fall into two entirely separate groups: loading a model, where the file itself can execute code on your machine, and running one, where it can be deceived or leak data. This chapter covers pickle, trust_remote_code, the limits of scanners, and a checklist before going live.
9. 09 Appendix — Vocabulary, Commands and a Learning Path The terms you will meet with short definitions, the everyday commands for checking a machine and managing models, a week-by-week learning path, and the whole guide condensed into eight lines.

This series is All Rights Reserved — free to read, but copying, downloading, or republishing requires written permission.