AI Essentials — free course, real certificate.

Artificial intelligence is software that performs tasks that normally require human intelligence — understanding language, recognising patterns, making predictions, generating content. No magic, no consciousness: maths, data and statistics at enormous scale.

Machine learning — the engine of modern AI

Traditional software follows rules a programmer wrote by hand. Machine learning flips that: the system learns patterns from examples in data. Show it millions of labelled emails and it learns what spam looks like — nobody writes a "spam rule" by hand.

Narrow vs general

Everything you can use today — chatbots, image generators, recommendation engines — is narrow AI: brilliant at specific tasks, with no general understanding of the world. Human-level "general AI" remains a research goal, not a product.

The mindset that matters

• AI output is probabilistic — a statistically likely answer, not a guaranteed fact.
• Treat it like a brilliant, fast, occasionally-wrong assistant — never an oracle.

Large language models (LLMs) — the engines behind modern AI assistants — are trained on vast amounts of text. From it they learn one deceptively simple skill: predicting the most likely next word (token). Generate one word, then the next, then the next — and fluent paragraphs, emails and code emerge. "Generative AI" simply means AI that creates new content this way: text, images, audio, code.

Four concepts worth knowing

• Training cutoff: a model's built-in knowledge comes from its training data, up to a cutoff date. By default it doesn't browse the web or check facts — it recalls patterns.
• Context window: how much text the model can consider at once — your conversation plus any documents you paste. Long inputs that exceed it get forgotten.
• Hallucination: because the model optimises for plausible, not true, it can state falsehoods fluently and confidently — invented references, wrong numbers, fake details. This is the single most important limitation to remember.
• Grounding (RAG): businesses connect models to their own documents — called retrieval-augmented generation — so answers draw on real, current company information instead of memory alone.

A prompt is just your instruction to the model — and the quality of what you get back tracks the quality of what you put in. The reliable formula:

Instruction + context + format

• Instruction: say exactly what you want done — "summarise", "rewrite for a customer", "list the risks".
• Context: who it's for, the situation, and any source material. Pasting the actual document beats describing it.
• Format: the shape of the answer — "three bullet points", "a 100-word email", "a table".

Power techniques

• Few-shot prompting: include one or two examples of the output you want — the model will match their style and structure.
• Step-by-step: for reasoning tasks, ask the model to work through the problem before answering.
• Ground it: when accuracy matters, provide the source text and instruct the model to use only that material — then verify.
• Iterate: treat the first response as a first draft. Refine, add constraints, ask again — that's the workflow, not a failure.

The fastest returns come from frequent, low-risk language tasks:

• Drafting: emails, proposals, job ads, social posts — start at 80% instead of 0%.
• Summarising: long reports, meeting transcripts, email threads — minutes become seconds.
• Rewriting: change tone, simplify jargon, tighten length, fix grammar.
• Extracting: pull names, dates, amounts and action items out of messy text.
• Brainstorming: a tireless first-ideas partner that never gets embarrassed.
• Frontline agents: AI receptionists and chat agents that answer, book and route 24/7 — increasingly common in Australian small business.

The two workplace rules

• Human-in-the-loop: AI drafts, a person reviews and approves before anything is sent, published or actioned. You own the output.
• Confidentiality: never paste sensitive company or customer information into tools your workplace hasn't approved — check the policy first.

• Hallucination: the headline risk. Fluent ≠ correct. Verify anything that matters — especially facts, figures, names and citations — before relying on it.
• Bias: models learn from human-made data, so they inherit its patterns and blind spots. Review outputs that touch people — hiring, lending, service decisions — with extra care.
• Privacy: information entered into consumer AI tools may be retained or used for training. Use approved tools, strip personal details where you can.
• Copyright: the law around AI-generated content is still settling. Don't pass off AI output as original creative work where it matters, and check ownership terms of the tools you use.
• Deepfakes & scams: voices and faces can now be convincingly faked. For any unusual or urgent request — especially payments — verify through a separate, trusted channel before acting.
• Disclosure: when AI played a significant role in something you deliver, be transparent about it. Trust is cheaper to keep than to win back.

• Week 1 — pick two tasks. Choose frequent, low-stakes work (drafting, summarising). Use AI on them daily, always reviewing output.
• Week 2 — build a prompt library. Save the prompts that work. A good prompt reused beats a great prompt forgotten.
• Week 3 — measure. Roughly track time saved and quality. Wins fund the case for going further; failures teach you the limits.
• Week 4 — make it official. Know (or help write) your workplace AI policy: which tools are approved, what data may be used, who reviews output, when to disclose.
• Always — keep current. Capability changes quarter to quarter. Re-test what "AI can't do" every few months, because the answer keeps changing.

That's the course. When you're ready, take the 20-question exam — pass with 80% and your certificate is issued on the spot.