RC
Rupert Chesman
or Space to navigate
1 / 138
Foundation Course

AI Fundamentals

Understanding AI properly — from prediction to prompting to safe, practical use.

Welcome

Introductions

Who are you, and what brought you here today?

"

This course is about understanding — not tool tutorials.

Once you understand how AI actually works, every tool becomes easier.

The philosophy behind AI Fundamentals

Agenda

Today's 7 Modules

  • Module 1 — What AI Is and Is Not
  • Module 2 — How LLMs Work
  • Module 3 — Models, Tools & Task Fit
  • Module 4 — Prompting & Context Design
  • Module 5 — Verification & Hallucinations
  • Module 6 — Safety, Privacy & Governance
  • Module 7 — Applied Adoption & Capstone

Who This Course Is For

  • Complete beginners who want to understand AI properly
  • Professionals who use AI tools but don't understand how they work
  • Managers who need to make decisions about AI adoption
  • Anyone who is "hype-fatigued" and wants the real picture
  • No technical background required — zero prerequisites

~21 hours

Total course length across 7 modules — we'll cover the highlights today

Module 1

What AI Is and Is Not

Building accurate mental models and separating capability from hype.

01

What AI Is and Is Not

Objectives

What You'll
Learn

  • Define AI, Machine Learning, and Generative AI in plain English
  • Distinguish AI from automation and regular software
  • Understand how models learn from data
  • Challenge overclaiming from marketing and media
Definition

AI in Plain English

A practical definition that actually makes sense.

What AI Actually Is

  • Artificial Intelligence: systems that perform tasks normally requiring human intelligence
  • Machine Learning: AI that learns patterns from data rather than following explicit rules
  • Deep Learning: ML using layered neural networks to find complex patterns
  • Generative AI: systems that create new content (text, images, code, audio, video)
  • Key insight: AI is narrow — it does specific tasks, not "everything"

What AI Is NOT

  • Not sentient, conscious, or self-aware
  • Not "thinking" in any human sense
  • Not a single general intelligence that can do anything
  • Not magic — it's mathematics and statistics at scale
  • Not infallible — it makes confident-sounding mistakes constantly
  • Not going to "take over" — it's a tool, like electricity
"

AI is autocomplete on steroids.

Incredibly useful. Not remotely sentient.

Distinction

AI vs Automation vs Software

When is something AI, and when is it just good programming?

The Difference

  • Regular Software: follows explicit rules (if X then Y)
  • Automation: repeats predefined steps without human intervention
  • Analytics: finds patterns in structured data using statistics
  • AI/ML: learns patterns from data and makes predictions on new inputs
  • Generative AI: creates new content based on learned patterns
  • Many things marketed as "AI" are actually just automation

Real-World Examples

  • AI:
  • ChatGPT writing an email
  • Image recognition in photos
  • Language translation
  • Medical image analysis
  • Fraud detection
  • NOT AI:
  • Email auto-sorting by rules
  • Spreadsheet formulas
  • Mail merge
  • Auto-responder emails
  • Password managers
Activity

Classify the Task

8 min

For each scenario below, decide: is it AI, automation, analytics, or regular software? 1. Spam filter that learns from your behaviour 2. Out-of-office reply 3. Netflix recommending shows 4. A pivot table in Excel 5. Siri answering a question 6. Scheduled social media posts

Learning

How AI Learns

Data in → patterns learned → predictions out.

Data, Training, and Inference

  • Data: the raw material AI learns from (text, images, numbers)
  • Parameters: the internal "knobs" the model adjusts during learning
  • Training: the process of adjusting parameters to match patterns in data
  • Inference: using the trained model to make predictions on new inputs
  • Key: the model finds statistical patterns — it doesn't "understand" meaning

A Simple Analogy

  • Imagine learning to predict the next word in a sentence
  • Read millions of sentences — notice which words follow which
  • "The cat sat on the ___" → "mat" is statistically likely
  • That's essentially what LLMs do — at massive scale
  • The result can seem intelligent, but it's pattern matching
  • Understanding this prevents overreliance and disappointment

Bias in Training Data

  • Models learn whatever is in their training data ��� including biases
  • If training data overrepresents certain viewpoints, so will outputs
  • Historical biases in text get absorbed and reproduced
  • This is why outputs can be racist, sexist, or culturally narrow
  • Not a "bug" that gets fixed — it's structural to how learning works
  • Always critical to review AI outputs through a human lens
Critical

Challenging AI Claims

Marketing says one thing. Reality often says another.

The AI Claims Fact-Check

  • Marketing: "Our AI understands your business"
  • Reality: It pattern-matches against training data
  • Marketing: "AI-powered insights"
  • Reality: Often just statistics with a chatbot front-end
  • Marketing: "Human-level reasoning"
  • Reality: Benchmark performance ≠ real-world reliability

How to Assess AI Claims

  • Look for: demos vs deployment (lab vs real world)
  • Look for: benchmark scores vs practical reliability
  • Look for: "powered by AI" that's actually just rules/automation
  • Look for: cherry-picked examples vs average performance
  • Ask: "What does this actually do better than the old way?"
  • Ask: "What happens when it gets it wrong?"
Activity

Spot the Overclaim

10 min

Find an AI product or service online (or think of one you've seen marketed). 1. What does the marketing claim? 2. What is it probably actually doing? 3. How would you verify the claim? Discuss with your neighbour.

Module 2

How LLMs Work

Prediction, tokens, context windows, and why outputs can be wrong.

02

How LLMs Work

Recap

What We
Covered

  • AI is pattern matching from data, not thinking
  • Difference between AI, automation, and software
  • Models learn from data — including its biases
  • Marketing claims often overstate real capability
Objectives

What You'll
Learn

  • Explain next-token prediction in plain English
  • Understand tokens and context windows
  • Know why the same prompt gives different outputs
  • Grasp the basics of transformer architecture
Core

Prediction, Not Understanding

The single most important thing to know about LLMs.

What LLMs Actually Do

  • LLM = Large Language Model
  • Core mechanic: predict the most likely next token
  • A "token" is roughly a word or part of a word
  • "The cat sat on the" → model predicts "mat" or "floor" or "sofa"
  • It does this billions of times to generate paragraphs
  • The output sounds intelligent but is statistical, not reasoned

Why This Matters

  • If you think AI "knows" things — you'll trust it too much
  • If you know it "predicts" things — you'll verify appropriately
  • It can be confidently wrong because confident language is statistically common
  • It doesn't "know" when it doesn't know something
  • Understanding prediction prevents overreliance
  • It also helps you write better prompts
"

It doesn't know. It predicts.

And sometimes those predictions are spectacularly wrong.

The key mental model for LLM use

Activity

Prediction Exercise

5 min

Complete these sentences — predict the next word: "To be or not to ___" "The quick brown fox jumps over the ___" "In a galaxy far, far ___" You just did what an LLM does — used patterns to predict. Now imagine doing that with billions of sentences of training data.

Mechanics

Tokens & Context Windows

How prompts are processed and why length matters.

What Are Tokens?

  • Models don't read words — they read tokens
  • A token is roughly 3/4 of a word (on average)
  • "Hello world" = 2 tokens
  • "Unbelievable" = 3 tokens (un + believ + able)
  • Numbers, punctuation, and spaces are also tokens
  • Why it matters: you pay per token, and context has limits

Context Windows

  • The context window is the model's "working memory"
  • It includes your prompt AND the response
  • GPT 5.5: ~128,000 tokens (~95,000 words)
  • Claude Opus 4.7: ~200,000 tokens (~150,000 words)
  • Gemini 3.1: ~2,000,000 tokens (~1.5M words)
  • When you exceed the window: older content gets truncated or ignored

Why Context Windows Matter

  • Longer context = more information the model can consider
  • But models don't use all context equally well
  • "Lost in the middle" — models pay more attention to start and end
  • Practical tip: put the most important info at the start
  • Practical tip: keep prompts focused rather than dumping everything in
  • More context isn't always better — clarity beats quantity
Variability

Temperature & Consistency

Why the same prompt gives different answers.

Why Outputs Vary

  • Temperature: a setting that controls randomness in predictions
  • Low temperature (0.0): very predictable, deterministic outputs
  • High temperature (1.0+): more creative, more varied outputs
  • Even at temperature 0, outputs can vary slightly between runs
  • This is by design — not a bug or failure
  • Use low temperature for facts, high for brainstorming

When Consistency Matters

  • Data extraction → low temperature (consistent formatting)
  • Classification tasks → low temperature (reliable categories)
  • Creative writing → higher temperature (more variety)
  • Brainstorming → higher temperature (divergent ideas)
  • Code generation → low temperature (predictable syntax)
  • If you need the same output every time, lower the temperature
Activity

Test Temperature

8 min

Open ChatGPT (or any model) and ask the exact same question three times: "Give me three team-building ideas for a team of 10 people" Are the results the same each time? What varies? What stays consistent?

Architecture

Transformers in Plain English

The architecture that made modern AI possible.

What Is a Transformer?

  • The architecture behind GPT, Claude, Gemini, and most modern LLMs
  • Key innovation: "attention" — the model can focus on relevant parts of input
  • Before transformers: models processed text word by word (slow, forgetful)
  • After transformers: models see the whole input at once and weigh importance
  • This is why modern models are so good at understanding context
  • You don't need to understand the maths — just the concept

Attention in Plain English

  • Imagine reading a sentence and highlighting the most important words
  • "The bank by the river was flooded" — "river" tells you which "bank"
  • Attention lets the model connect related words across the whole input
  • This is why context and clear prompts matter so much
  • The model weighs: "what parts of this input are most relevant to generating the next word?"
  • Better context = better attention = better outputs
"

You don't need to understand the engine to drive the car.

But knowing it runs on fuel, not magic, changes how you use it.

Module 3

Models, Tools & Task Fit

Choosing the right model and tool for the job.

03

Models, Tools & Task Fit

Recap

What We
Covered

  • LLMs predict likely next tokens — they don't "know" things
  • Tokens and context windows determine what models can process
  • Temperature controls randomness in outputs
  • Transformers use attention to focus on relevant context
Objectives

What You'll
Learn

  • Compare ChatGPT, Claude, Gemini, and open models
  • Choose the right model for different tasks
  • Know when NOT to use AI
  • Build a personal starter tool stack
Landscape

The Major Models

Choosing by task, not brand loyalty.

Proprietary vs Open Models

  • Proprietary (Hosted):
  • ChatGPT (OpenAI)
  • Claude (Anthropic)
  • Gemini (Google)
  • Grok (xAI)
  • Hosted for you, pay per use
  • Open (Downloadable):
  • Gemma (Google)
  • Mistral
  • DeepSeek
  • Run locally, free to use

ChatGPT 5.5 (OpenAI)

  • The most popular — largest ecosystem of apps built on it
  • December 2025 knowledge cutoff
  • Strengths: general tasks, writing, analysis, image creation
  • Weaknesses: can be sanitised, corporate tone
  • Features: Canvas, Agent Mode, Deep Research, Custom GPTs
  • Best for: everyday tasks, writing, quick image generation

Claude Opus 4.7 (Anthropic)

  • Currently the strongest model for coding and reasoning
  • January 2026 knowledge cutoff
  • Strengths: nuanced reasoning, long documents, coding
  • Weaknesses: slower for simple tasks, fewer integrations
  • Features: Cowork (computer control), Projects, long context
  • Best for: complex analysis, coding, document review

Gemini 3.1 (Google)

  • Google's model — deeply integrated with Google ecosystem
  • Strengths: research, multimodal, massive context (2M tokens)
  • Weaknesses: sometimes inconsistent quality
  • Features: NotebookLM, Google AI Studio, Google integration
  • Best for: research, working with Google tools, long documents
  • Gemini 3.1 Flash: fast, cheap variant for simple tasks

Grok 4.20 (xAI)

  • Built on X (Twitter) data — always current
  • The only model with no knowledge cutoff
  • Strengths: current events, creative, fewer guardrails
  • Weaknesses: can be inappropriate, less corporate-friendly
  • Best for: current events, creative work, unfiltered analysis
  • Not recommended for professional/corporate output

Open Models

  • Models you can download and run on your own hardware
  • Gemma 4 (Google): small but capable, runs on modest hardware
  • DeepSeek V4: strong reasoning, open-source roots
  • Advantages: privacy (data never leaves your machine), free, customisable
  • Disadvantages: need technical setup, less powerful than frontier models
Choice

Task-Model Fit

Choose by what you need done, not by brand.

Matching Tasks to Models

  • Quick writing/editing → ChatGPT (fast, good enough)
  • Complex analysis → Claude (best reasoning)
  • Research with sources → Gemini (grounded in search)
  • Current events → Grok (no knowledge cutoff)
  • Privacy-sensitive → Open models (runs locally)
  • Code generation → Claude or ChatGPT (both strong)

When NOT to Use AI

  • High-stakes legal or medical decisions without expert review
  • Tasks where a wrong answer could cause real harm
  • Confidential data you can't risk being stored/trained on
  • Simple tasks where a Google search is faster and more reliable
  • When you need 100% accuracy (AI gives ~95% at best)
  • When the verification cost exceeds the generation value
Activity

Build Your Starter Stack

10 min

Based on what you've learned, create a personal "AI Starter Stack": 1. Primary tool: (for everyday tasks) 2. Secondary tool: (for specific needs) 3. Rules: (what you WILL and WON'T use AI for) Write it on paper or in a note — this is your v1 policy.

Module 4

Prompting & Context Design

Writing clear prompts that get useful results.

04

Prompting & Context Design

Recap

What We
Covered

  • Different models suit different tasks
  • ChatGPT for general, Claude for reasoning, Gemini for research
  • Open models for privacy, Grok for current events
  • Know when NOT to use AI
Objectives

What You'll
Learn

  • Write clear, effective prompts
  • Use examples and role prompting
  • Apply prompting across work tasks
  • Chain prompts for complex workflows
Foundation

Writing Clear Prompts

The single most useful skill in AI.

Anatomy of a Good Prompt

  • Goal: What do you want the AI to produce?
  • Context: What background does it need?
  • Constraints: What are the limits (length, format, tone)?
  • Output format: How should the result look?
  • Audience: Who is this for?
  • The clearer you are, the better the output

The RCTF Framework

  • Role: Who should the AI be? (domain expert, editor, analyst)
  • Context: What's your situation? (project, audience, constraints)
  • Task: What exactly do you need done?
  • Format: How should the output be structured?
  • This framework works with every model, every time
  • Example: "Act as a marketing strategist. I run a small bakery in Sydney. Create a 7-day social media plan. Format as a table."

Bad Prompt vs Good Prompt

BAD: "Write me a marketing plan" GOOD: "Act as a digital marketing strategist with 10 years experience in hospitality. I run a 40-seat cafe in Newtown, Sydney. We're strong on weekdays but quiet on weekends. Our target audience is 25-40 year olds who work from home. Please create a 4-week social media plan focused on increasing weekend foot traffic. Format as a weekly calendar with specific post ideas, captions, and best posting times for Instagram and Facebook."

Activity

Rewrite Vague Prompts

10 min

Rewrite each of these vague prompts into effective ones using RCTF: 1. "Help me with my presentation" 2. "Write an email" 3. "Give me ideas" 4. "Summarise this document" Share your best rewrite with the group.

Technique

Examples & Role Prompting

Show the AI what you want — don't just tell it.

Few-Shot Prompting

  • Give the AI examples of what you want
  • One example: "one-shot" prompting
  • Multiple examples: "few-shot" prompting
  • The model mimics the pattern you demonstrate
  • Especially useful for: consistent formatting, specific tone, classification
  • Show 2-3 good examples → get consistent results

Few-Shot Example

Classify these customer enquiries as Sales, Support, or Billing: Examples: "How much does the Pro plan cost?" → Sales "My login isn't working" → Support "I was charged twice this month" → Billing Now classify: "Can I upgrade from the Basic plan?" "The export button gives an error" "Do you offer team discounts?"

Role Prompting

  • Tell the AI WHO to be — it changes everything
  • "Act as a lawyer" vs "Act as a copywriter" → totally different responses
  • Stack roles: "Act as both a data analyst AND a storyteller"
  • Include experience: "...with 15 years in Australian retail"
  • The role primes domain-specific vocabulary and reasoning
  • It's not pretending — it's activating relevant patterns

Output Formatting

  • Don't leave format to chance — specify exactly what you want
  • "Format as a table with columns: Date, Task, Owner, Status"
  • "Return JSON with keys: name, category, priority"
  • "Use bullet points, keep each under 20 words"
  • "Write in the style of a BBC news article"
  • Explicit format instructions dramatically improve usefulness
Applied

Prompting for Work Tasks

Templates you can use immediately.

Common Work Prompts

  • Summarising: "Summarise this in 3 bullet points for a busy executive"
  • Analysing: "List the top 5 risks in this proposal and rate each 1-5"
  • Planning: "Create a project plan with milestones for the next 90 days"
  • Editing: "Rewrite this email to be more concise and professional"
  • Ideation: "Give me 10 different approaches to this problem"
  • Each of these becomes more powerful with RCTF context
Activity

Build a Prompt Pack

15 min

Think about your actual job. Create 3 reusable prompt templates for tasks you do regularly: 1. A task you do daily 2. A task you do weekly 3. A task that takes too long manually Use RCTF structure. Keep them — these are your AI shortcuts.

Advanced

Prompt Chaining

Breaking complex tasks into steps.

What Is Prompt Chaining?

  • Breaking one big task into a sequence of smaller prompts
  • Each prompt builds on the output of the previous one
  • Step 1: Generate ideas → Step 2: Evaluate ideas → Step 3: Expand best idea
  • Like a production line — each station does one thing well
  • Results are much better than asking for everything at once
  • Also easier to fix when something goes wrong

Chaining Example: Blog Post

Step 1: "Give me 5 blog post angles about [topic] for [audience]" Step 2: "Expand angle #3 into a detailed outline with H2 headings" Step 3: "Write the introduction paragraph — hook the reader in 3 sentences" Step 4: "Now write section 2, keeping the tone conversational and under 200 words" Each step is small, focused, and easy to redirect.

Context Packaging

  • When a task needs background info, package it cleanly
  • "Here is the project brief: [paste]"
  • "Here are the constraints: [list]"
  • "Here is an example of what good looks like: [paste]"
  • Keep context focused — don't dump everything
  • Label sections clearly so the model knows what's what
Activity

Chain a Task

12 min

Take one complex task from your work and break it into a 3-4 step prompt chain: 1. What info do you need first? 2. What decision comes next? 3. What's the final output? Test it in your preferred AI tool.

Module 5

Verification & Hallucinations

Checking outputs, reducing errors, and spotting failure modes.

05

Verification & Hallucinations

Recap

What We
Covered

  • RCTF framework: Role, Context, Task, Format
  • Few-shot examples improve consistency
  • Prompt chaining breaks complex tasks into steps
  • Explicit formatting dramatically improves outputs
Objectives

What You'll
Learn

  • Explain what hallucinations are and why they happen
  • Use verification workflows to check AI outputs
  • Design simple evaluation criteria
  • Practice spotting errors in AI-generated content
Critical

What Hallucinations Are

Confident-sounding inaccuracies — not bugs, not lies.

Understanding Hallucinations

  • When AI generates information that sounds correct but is fabricated
  • NOT lying (requires intent) — NOT a bug (it's working as designed)
  • It's the natural result of prediction without understanding
  • Common examples: fake citations, invented statistics, wrong dates
  • Made-up people, fictional court cases, non-existent research papers
  • The model doesn't "know" when it's wrong — it just predicts

When Hallucinations Are Most Likely

  • Obscure topics with limited training data
  • Very specific factual claims (dates, numbers, names)
  • When asked to cite sources (often invents plausible-looking URLs)
  • When the model is forced to give an answer it shouldn't
  • Under time pressure or long conversations
  • When the prompt is ambiguous or contradictory

Every AI model
hallucinations

No model is immune — verification is always your responsibility

Types of Hallucination

  • Fabrication: completely made-up facts presented as true
  • Conflation: mixing up real facts from different contexts
  • Extrapolation: extending patterns beyond what's supported
  • Citation hallucination: inventing sources that don't exist
  • Temporal confusion: misattributing events to wrong dates
  • All of these sound equally confident — that's the danger
Activity

Catch the Hallucination

10 min

Ask your AI tool a specific factual question about something you know well: - A specific statistic from your industry - A date/fact about your organisation - Details of a real event you attended Did it get it right? If wrong, how confident did it sound?

Practice

Verification Workflows

Practical habits for checking AI outputs.

The Verification Checklist

  • 1. Source check: Can I verify this claim independently?
  • 2. Consistency: Does this match other known facts?
  • 3. Specificity check: Are numbers/dates/names actually real?
  • 4. Citation check: Do referenced sources actually exist?
  • 5. Logic check: Does the reasoning actually follow?
  • 6. Bias check: Is this presenting one side as universal?

Verification Strategies

  • Cross-reference: check claims against a second source
  • Ask the model: "Are you certain? Can you verify this?"
  • Upload source material: give it the facts rather than asking for them
  • Use grounded search: Gemini with Google Search, Perplexity
  • For critical work: always verify independently of the model
  • Rule of thumb: the more specific a claim, the more you should check

When to Trust vs Verify

  • Safe to trust: creative brainstorming, general explanations, formatting
  • Verify lightly: summaries of your own documents, code logic
  • Verify thoroughly: specific facts, statistics, legal/medical claims
  • Never trust blindly: citations, financial figures, people's names
  • Key principle: use AI for the first draft, your brain for verification
  • The 80/20 rule: AI does 80% of the work, you verify the crucial 20%
Activity

Build a Verification Habit

10 min

Take an AI-generated response you've received today and run it through the verification checklist: 1. Can you verify the key claims? 2. Are any numbers/dates/names accurate? 3. Do the cited sources (if any) actually exist? Score: how many checks passed?

Evaluation

Evaluation for Normal People

Simple rubrics — not complex engineering.

Defining "Good Enough"

  • Before using AI, decide: what does a good output look like?
  • Set simple success criteria: accuracy, completeness, tone, format
  • Create test cases: try the same prompt 3-5 times
  • Check: are results consistent? Are they useful? Are they safe?
  • If results vary wildly, the prompt needs more structure
  • Evaluation doesn't need to be complex — just intentional

Your Personal Evaluation Framework

  • For each recurring AI task, define:
  • ✓ What must be accurate (non-negotiable)
  • ✓ What must be complete (all required elements)
  • ✓ What tone/style is needed
  • ✓ What format it should follow
  • ✓ What errors would be unacceptable
  • Write these down — they become your quality standard
Activity

Design an Eval

8 min

Pick one recurring task you plan to use AI for. Define: 1. Three things that must be accurate 2. The required format 3. One thing that would make the output unacceptable This is your evaluation rubric for that task.

Module 6

Safety, Privacy & Governance

Using AI responsibly, safely, and ethically.

06

Safety, Privacy & Governance

Recap

What We
Covered

  • Hallucinations are confident-sounding inaccuracies
  • Verification is always your responsibility
  • Cross-reference, check citations, verify specifics
  • Define success criteria before using AI for recurring tasks
Objectives

What You'll
Learn

  • Understand responsible use and potential harms
  • Know what data to never share with AI
  • Handle copyright and attribution properly
  • Create a lightweight AI usage policy
Ethics

Responsible Use

AI can help or harm — the choice is in how we use it.

Potential Benefits

  • Democratising access to expertise and knowledge
  • Reducing time on repetitive tasks
  • Making complex information more accessible
  • Supporting creativity and ideation
  • Enabling people with disabilities
  • Accelerating research and discovery

Potential Harms

  • Spreading misinformation at scale
  • Reinforcing and amplifying biases
  • Displacing workers without adequate transition support
  • Privacy violations through data collection/training
  • Environmental cost of training and running models
  • Enabling manipulation, fraud, and deception

Your Responsibility

  • The person using AI is responsible for its outputs
  • "The AI made it" is not a valid excuse for harmful content
  • You must verify before sharing or publishing
  • You must consider who might be affected
  • If something feels wrong, stop and think
  • Ethical AI use is everyone's job, not just a tech team's
Privacy

Privacy & Confidentiality

What should and should not go into AI systems.

Data You Should NEVER Paste Into AI

  • Passwords, API keys, access tokens
  • Personal data: names + addresses + financial info of others
  • Medical records or health information
  • Confidential business strategy or unreleased financials
  • Legal documents under privilege
  • Client data without explicit consent

How AI Tools Handle Your Data

  • By default, most tools use your inputs to train future models
  • ChatGPT: Settings → Data Controls → Turn off model training
  • Gemini: Activity → Turn off Gemini Apps Activity
  • Enterprise plans: typically don't train on your data
  • Key question: "If this leaked, would it matter?"
  • If yes → don't paste it, or use enterprise-grade tools

Safe Practices

  • Anonymise data before sharing with AI
  • Use placeholder names and numbers for sensitive work
  • Prefer enterprise/business tiers for professional use
  • Read the terms of service — know what they do with your data
  • Don't share screenshots that contain confidential info
  • Treat AI tools like a helpful stranger — not a trusted colleague
Activity

Data Audit

8 min

Think about the last 5 things you put into an AI tool: 1. Did any of them contain personal data? 2. Did any contain confidential business info? 3. Is model training turned off in your settings? Check your settings now. Turn off training if needed.

Legal

Copyright & Attribution

What you can and can't do with AI-generated content.

Copyright Basics for AI

  • AI-generated content has unclear copyright status in most jurisdictions
  • In Australia: no clear legal precedent yet for pure AI output
  • Safe approach: treat AI output as a draft that you then transform
  • Add substantial human creativity, editing, and judgement
  • Never pass off AI content as 100% human-written in contexts that require it
  • When using AI on others' content: the same copyright rules still apply

Attribution Best Practices

  • Be transparent about AI assistance where it matters
  • Academic work: always disclose AI use per institution policy
  • Professional work: follow your organisation's disclosure rules
  • Marketing: don't claim expertise that was actually AI-generated
  • Images: clarify when imagery is AI-generated
  • General rule: if someone would feel misled, disclose
Governance

Governance for Individuals & Teams

Lightweight rules that keep you safe.

The Green / Amber / Red Framework

  • GREEN: Safe to use AI freely
  • → Brainstorming, drafting, internal notes, learning
  • AMBER: Use AI with human review
  • → Client-facing content, reports, external communications
  • RED: Do not use AI (or use with extreme caution)
  • → Legal advice, medical decisions, confidential data, HR decisions

Creating Your AI Policy

  • Every team should have a one-page AI usage policy
  • Include: approved tools, data handling rules, review requirements
  • Include: disclosure expectations, escalation process
  • Include: what's explicitly prohibited
  • Review quarterly — AI capabilities change fast
  • Start simple, add complexity only where needed
Activity

Draft Your Policy

15 min

Create a one-page AI usage policy for your team or yourself: 1. List 3 GREEN tasks (free to use AI) 2. List 3 AMBER tasks (AI + human review) 3. List 3 RED tasks (no AI or extreme caution) 4. One rule about data handling 5. One rule about disclosure Keep it simple and practical.

Module 7

Applied Adoption & Capstone

Turning understanding into daily practice.

07

Applied Adoption & Capstone

Recap

What We
Covered

  • AI use carries responsibility for outputs
  • Never paste confidential data into consumer AI tools
  • Copyright for AI output is still legally unclear
  • Green/Amber/Red framework for usage governance
Objectives

What You'll
Learn

  • Build a personal AI operating system
  • Plan team adoption responsibly
  • Complete a real-world capstone task
  • Choose your next learning pathway
Personal

Your Personal AI Operating System

Two or three high-value use cases that save real time.

Finding Your High-Value Use Cases

  • Look for tasks that are: repetitive, text-heavy, time-consuming
  • Look for tasks where: 90% accuracy is good enough
  • Look for tasks where: you can verify the output easily
  • Common winners: email drafting, meeting prep, research summaries
  • Common winners: content ideation, data formatting, learning new topics
  • Start with ONE task. Master it. Then add more.

Your Weekly AI Map

  • Monday: plan the week using AI to draft priorities
  • Daily: use AI for email drafting and meeting preparation
  • Wednesday: AI-assisted content creation or report drafting
  • Friday: use AI to summarise the week's learnings
  • Ongoing: AI as a thinking partner for problem-solving
  • This isn't about replacing yourself — it's about augmenting
Activity

Map Your Week

12 min

Map your actual work week. For each day: 1. What task takes the most time? 2. Could AI help with any part of it? 3. What would verification look like? Pick your top 2-3 use cases. These become your personal AI operating system.

Team

Team Adoption

Helping others adopt AI without forcing hype-led rollouts.

Responsible Team Adoption

  • Start small: one team, one use case, one tool
  • Get buy-in: involve the team in choosing the use case
  • Set clear rules: what's allowed, what needs review, what's off-limits
  • Pilot first: 2-4 weeks with defined success criteria
  • Review: did it actually save time? Were there issues?
  • Scale: only expand what's proven to work

The Team Pilot Brief

  • Use case: What specific task are we automating/augmenting?
  • Tool: Which AI tool will we use?
  • People: Who is involved?
  • Duration: How long is the pilot?
  • Success criteria: How do we know it worked?
  • Guardrails: What data/tasks are excluded?
  • Review date: When do we assess and decide?
Activity

Design a Pilot

15 min

Draft a team AI pilot brief: 1. Choose one task your team does repeatedly 2. Choose a tool 3. Define success (how will you know it worked?) 4. Set guardrails (what's off-limits?) 5. Set a timeline (2-4 weeks) Present your pilot to the group.

Capstone

Capstone Project

Solve a real task using everything you've learned.

Capstone Requirements

  • Choose a REAL task from your work or life
  • Use at least one AI tool with proper prompting
  • Apply RCTF framework in your prompts
  • Verify the output using your verification checklist
  • Document what worked, what didn't, and what you'd change
  • Present your solution to the group (3-5 minutes)
Activity

Capstone: Solve Something Real

25 min

Choose a real task and complete it end-to-end with AI: Requirements: ✓ Use RCTF prompting ✓ Chain at least 2 prompts ✓ Verify the output ✓ Note what you'd improve Ideas: write a proposal, analyse a document, create a plan, draft communications, solve a problem.

Activity

Present Your Capstone

20 min

Present your capstone to the group (3-5 min): 1. What task did you solve? 2. What tool(s) did you use? 3. How did you prompt it? 4. How did you verify the output? 5. What would you do differently?

Next

What Comes Next

Your pathway from here.

Your Immediate Action Items

  • Today: Set up RCTF as your default prompting framework
  • Today: Turn off model training in your AI tool settings
  • This week: Use AI for one real task using what you learned
  • This week: Run verification on every AI output before sharing
  • This month: Build your personal AI operating system (2-3 use cases)
  • This month: Share the Green/Amber/Red framework with your team

Recommended Next Courses

  • Mastering AI Tools — deep dive into specific tools and workflows
  • AI for Corporate Teams — AI adoption strategy and governance
  • AI Agents & Automation — building workflows and autonomous systems
  • Vibe Coding — creating apps by describing what you want
  • AI for Creatives — image, video, and content creation
  • Visit www.rupertchesman.com for all courses

Resources

  • Course materials: www.rupertchesman.com
  • AI Prompt Builder: www.rupertchesman.com/tools/prompt-builder
  • Hallucination Spotter: www.rupertchesman.com/tools/hallucination-spotter
  • Cheat sheets: www.rupertchesman.com/cheatsheets
  • All resources: www.rupertchesman.com/resources

Certificate Pathway

Complete all modules + capstone + quiz = AI Fundamentals Certificate

Q&A

Questions

What would you like to know more about?

Thank You

Thank You

www.rupertchesman.com • © Rupert Chesman 2026