LLM101 · How Large Language Models Work
About Course
A clear, non-mathematical introduction to how large language models are trained, how they turn text into tokens and predictions, and where their strengths and limits come from.
Who it is for
Professionals and beginning developers who want a solid mental model of LLMs before building applications with them.
Prerequisites
No prior programming or machine learning experience is required.
What you will be able to do
- Describe the steps of pretraining, fine-tuning and preference tuning in plain language
- Estimate token counts and context-window usage for a realistic business document
- Explain how temperature and sampling settings change model output
- Identify at least three failure modes, such as hallucination and outdated knowledge, and a mitigation for each
Lessons
- Lesson 1 · From text to tokens to predictions
- Lesson 2 · How models are trained and tuned
- Lesson 3 · Context, sampling and limits
- Lesson 4 · Lab and assignment
Lab: Tokens, context and sampling in practice
You will tokenize real business text, compare token counts across languages, and observe how a small open model's next-token probabilities change with temperature. The goal is to see the mechanics described in the lessons with your own eyes, not to build an application yet.
Free tools: Google Colab (free) with the open-source tiktoken library and a small open model from Hugging Face
Assessed assignment
Write a two-page explainer called "How our AI assistant actually works" for a non-technical small-business owner, such as a dental clinic manager. Explain tokens, context windows, training stages and sampling using examples from your lab notebook. Include one table that lists three failure modes, a realistic example of each in that business, and a practical mitigation. Keep the tone accurate and avoid hype.
| Criterion | Weight | What excellent work looks like |
|---|---|---|
| Technical accuracy | 35% | Every explanation of tokens, training and sampling is correct and matches what the lab showed. |
| Clarity for a non-technical reader | 25% | A business owner could follow the explainer without a glossary; analogies are helpful and not misleading. |
| Failure modes and mitigations | 25% | Three distinct failure modes with realistic examples and mitigations that a small business could actually apply. |
| Use of lab evidence | 15% | Token counts, probabilities or sample outputs from the notebook are cited to support the points made. |
Portfolio piece
A plain-language explainer on how LLMs work plus the Colab notebook that backs it up with real token counts and sampling experiments.
Next step
Continue with LLM102 · Python Essentials for AI Applications · 4 hours · $28.00
Course Content
LLM101 · How Large Language Models Work (4 h)
-
Lesson 1 · From text to tokens to predictions
-
Lesson 2 · How models are trained and tuned
-
Lesson 3 · Context, sampling and limits
-
Lesson 4 · Lab and assignment
-
LLM101 Assessment · How Large Language Models Work