Module 02: Claude Fundamentals¶
Level: Beginner · Time: ~4 hours including exercises · Exam relevance: CCAO-F D1, D5, D6 · CCAR-P D2, D5
Module 01 explained language models in general. This module is about Claude specifically: how Anthropic trains it, what happens when you send it a message, how to hold a productive first conversation, the basics of a good prompt, and how Anthropic and you share responsibility for using it safely.
If you have never used Claude before, this is where you start doing rather than reading.
Learning objectives¶
After completing this module you will be able to:
- Explain at a conceptual level how Claude is trained, including pre-training, reinforcement learning from human feedback (RLHF) and Constitutional AI.
- Explain why Claude conversations are stateless, and how the system, user and assistant roles fit together.
- Start a conversation in a Claude app, attach a file, iterate on a response and know when to start a new chat.
- Write a prompt that states the task, context, audience, format and constraints, and improve it based on the result.
- Describe Anthropic's safety approach: the Usage Policy, the Responsible Scaling Policy and Claude's constitution.
- Apply responsible-use practices: data sensitivity, verification, disclosure and following organisational AI policy.
Page order¶
| # | Page | What you learn | Time |
|---|---|---|---|
| 1 | How Claude works | Training, statelessness and conversation roles | 40 min |
| 2 | Your first conversation | A guided walk-through in the Claude app | 40 min |
| 3 | Prompt fundamentals | The building blocks of a clear request | 40 min |
| 4 | Anthropic's safety approach | Usage Policy, Responsible Scaling Policy, constitution | 35 min |
| 5 | Responsible use | Your responsibilities as a user or builder | 35 min |
Prerequisites¶
- Module 01: AI and LLM fundamentals, or equivalent familiarity with tokens, context windows and hallucinations.
- A Claude account for the hands-on parts (a free account is enough). If you cannot create one, you can still follow along and run the labs offline in mock mode; see Environment setup.
Exam relevance¶
Exam facts (item counts, timing, weights) are recorded in registry.yaml.
| Page | CCAO-F objectives | CCAR-P objectives |
|---|---|---|
| 01 How Claude works | CCAO-F-3.4, CCAO-F-5.3 | CCAR-P-2.2 |
| 02 First conversation | CCAO-F-1.3, CCAO-F-3.1, CCAO-F-3.4 | — |
| 03 Prompt fundamentals | CCAO-F-1.1, CCAO-F-1.2, CCAO-F-1.3, CCAO-F-1.4 | CCAR-P-2.2, CCAR-P-2.3 |
| 04 Anthropic's safety approach | CCAO-F-6.1, CCAO-F-6.4 | CCAR-P-5.1, CCAR-P-5.5 |
| 05 Responsible use | CCAO-F-6.1, CCAO-F-6.2, CCAO-F-6.3, CCAO-F-6.4, CCAO-F-2.4 | CCAR-P-5.3, CCAR-P-5.4 |
Practice¶
- Lab 01: First Claude interaction: send a message with system and user roles (mock mode by default).
- Lab 02: Prompt improvement: improve a weak prompt step by step.
- Lab 15: Sensitive data protection: an advanced lab on keeping sensitive data out of prompts; skim the README now and return later.
Next module¶
Business users heading for CCAO-F can continue to Module 03: Prompt engineering and then Module 05: Claude products and models. Everyone should complete Module 03 before Module 05.