How AI Really Works: From Prompt to Token
A technical deep-dive into what actually happens between typing a prompt into an AI system and getting a response back.
From the 70-year history of chatbots through tokenization, embeddings, transformers, and how a base model becomes a helpful assistant.
Learning Objectives
- Understand what a large language model actually is and how it differs from earlier chatbots
- Trace the technical lineage from statistical language modelling to modern transformers
- Explain tokenization, embeddings, and self-attention in plain language
- Describe how a base model becomes an instruction-following assistant via fine-tuning, RLHF, and tool calling
Prerequisites
No formal prerequisites — this course is written for anyone curious about how AI actually works.
- Helpful, but not required: comfort reading basic Python code (each module includes short, real code examples)
- Helpful, but not required: general familiarity with using an AI chatbot like Claude or ChatGPT
Course Content
Module 1: The Origins of AI & Chatbots
Before large language models, six decades of rule-based chatbots and one landmark experiment in information theory set the stage for everything that followed.
Large Language Models (LLM) - How they Work
1.1 What Is a Language Model, Really?
1.2 Claude Shannon and the Predictability of Language
1.3 The Rule-Based Chatbot Era (1966–2001)
1.4 Module 1 Quiz — Origins of AI & Chatbots
Module 2: How Neural Networks Learn
Neural networks are the engine inside every modern AI system. This module traces the idea from a 1943 theoretical paper, through a hyped-up 1958 machine, a near-fatal 1969 critique, and the 1986 breakthrough that made today's deep learning possible.
A Neural Network: What it is.
2.1 What Is a Neural Network?
2.2 From McCulloch-Pitts to the Perceptron (1943–1958)
PT2: Gradient descent, how neural networks learn | Deep Learning
2.3 The XOR Problem and the Rise of Backpropagation (1969–1986)
PT3:Deep Learning | Backpropagation, intuitively
2.4 Module 2 Quiz — How Neural Networks Learn
Module 3: Tokenization & Embeddings
Before a neural network can process a prompt, that prompt has to become numbers. This module covers how text is split into tokens, where the byte pair encoding algorithm actually came from, and how embeddings give a model a numeric sense of meaning.
3.1 What Is Tokenization?
3.2 Byte Pair Encoding: From Compression Trick to Language Vocabulary
3.3 Word Embeddings and the Geometry of Meaning
Embeddings
3.4 Module 3 Quiz — Tokenization & Embeddings
Module 4: Inside the Transformer
The Transformer architecture powers essentially every modern large language model. This module covers where it came from, how self-attention lets tokens exchange information, how a model actually chooses its next word, and how it all fits inside a context window.
Transformers
4.1 The Transformer Architecture: An Overview
4.2 Self-Attention: How Tokens Talk to Each Other
4.3 From Logits to Language: Softmax, Temperature, and Sampling
4.4 Context Windows and the Full Picture
4.5 Module 4 Quiz — Inside the Transformer
Module 5: From Autocomplete to Assistant
Everything so far describes a base model: a very capable next-token predictor with no particular tendency to be helpful. This final content module covers fine-tuning, reinforcement learning from human feedback, and tool calling — the steps that turn that base model into an assistant.
5.1 Pretraining, Fine-Tuning, and the Base Model
5.2 Reinforcement Learning from Human Feedback (RLHF)
RLHF
5.3 Tool Calling: Letting Models Act in the World
Tool Calling
5.4 Module 5 Quiz — From Autocomplete to Assistant
Module 6: Course Assignment
The capstone for this course: demonstrate, in your own words, that you can trace a prompt through the entire pipeline covered across Modules 1 through 5.
Large Language Models (Revision)
6.1 Capstone Assignment: Trace a Prompt Through the Full Pipeline
Premium Content Preview
This course includes:
- 5 interactive quizzes
- 31 comprehensive lessons
- Multiple lesson formats (videos, readings, activities)
- Progress tracking and certificates
Enroll now to access all content!
Course Information
Need Help with This Course?
Have a question about How AI Really Works: From Prompt to Token? Message Michael Williams directly and we'll get back to you.