KnowledgeInSight
AI Literacy
0% of Course 1 complete

Course 1 of 3

How LLMs Work

This course explains what is happening when an AI model answers you. You'll follow the path from machine learning basics to the training of large language models, then to the steps that turn a trained model into an assistant. You'll see why models produce fluent errors, how reasoning, tools, and agents extend what they can do, and how AI systems are now used in AI research itself. Everything is explained in words, with no math or code.

3 modules · 9 lessons · 88 items · 12h 11m

Start Course 1

Who this is for

You can hold a conversation with an AI chatbot and want to know what is going on behind the answers. You don't need any technical background.

What you will learn

  • Explain how a language model learns from text and generates a response one token at a time
  • Describe how pretraining, fine-tuning, and system instructions shape an assistant's behavior
  • Explain why models err, and how reasoning, tools, and agents extend what they can do
  • Read a benchmark or capability claim and say what it does and doesn't show

Contents

  1. Module 1 Machine Learning and Language Models 3 lessons

    This module covers the ideas underneath every AI chatbot: learning from examples, neural networks, and predicting the next piece of text. You'll be able to explain how a system can acquire an ability nobody programmed into it and how it writes a reply one token at a time. These ideas are worth having on their own, because most claims about what AI "knows" or "thinks" make sense only once you know what training and prediction are.

    Module overview and further reading

    1. 1.1Machine Learning Basics: Learning from examples instead of rules1h 3m
    2. 1.2Machine Learning Basics: Neural networks and how training adjusts them1h 3m
    3. 1.3Machine Learning Basics: Tokens, meaning, and next-token prediction1h 33m
  2. Module 2 Building a Modern Model 3 lessons

    This module follows a large language model from design to finished assistant. You'll be able to explain what the transformer changed, what pretraining consumes and leaves behind, and how a raw model is shaped into an assistant with a recognizable manner. Knowing these stages lets you tell which of a chatbot's traits come from its training data, which from its developer's choices, and which from the instructions around your conversation.

    Module overview and further reading

    1. 2.1Building a Model: The transformer and attention1h 3m
    2. 2.2Building a Model: Pretraining at scale: data, compute, and cost1h 3m
    3. 2.3Building a Model: From pretrained model to assistant1h 37m
  3. Module 3 Capabilities, Limits, and Trajectory 3 lessons

    This module looks at what current models can and can't do and how anyone knows. You'll be able to explain why a model states falsehoods fluently, how reasoning, search, tools, and agents extend it, and how to read a claim about AI progress, including claims about AI systems doing AI research. This is the knowledge you need to weigh a capability headline on your own.

    Module overview and further reading

    1. 3.1Capabilities and Limits: Why models err, and what is known about their inner workings1h 3m
    2. 3.2Capabilities and Limits: Reasoning, tools, and agents1h 3m
    3. 3.3Capabilities and Limits: Measuring progress and AI-assisted AI research1h 33m

Course project

Explain a Model's Answer You'll take one real reply from an AI assistant and write a plain-language account of how it came to be, from training through to the words on your screen. It's worth doing because explaining one concrete case shows you what you understand and where your… 3 items · 1h 10m