AI Foundations

Six words that get used interchangeably in the news and mean different things. Learn them once and most AI headlines become readable.

• Tell the difference between AI, machine learning and generative AI
• Explain what a large language model is in one sentence
• Recognise when a headline is describing something other than what you use

A working mental model of what today's AI tools do, why they sound confident when they are wrong, and what that means for how you use them.

• Explain in one sentence what a large language model does
• Predict the kinds of task it will be good and bad at
• Recognise why a wrong answer arrives in the same confident tone as a right one

A working list of the tasks these tools handle well, the ones they handle badly, and the ones that look identical from outside but are not.

• Predict which tasks a language model will handle well before you spend time on it
• Explain why arithmetic and counting are unexpectedly weak
• Identify the tasks that need care rather than a straight yes or no

Why AI seemed to appear overnight when it had been around for decades, and what that history tells you about how to keep learning.

• Explain what changed around 2022 and why it felt sudden
• Describe why capability keeps moving and what that means for your own learning
• Set a personal benchmark you can retest as tools change

Four kinds of AI assistant, what each is for, and why learning the categories outlasts learning the products.

• Name the four categories of assistant and what each is best at
• Explain why a search-connected tool and an offline one behave differently
• Choose a category before choosing a product

What the paid tiers add, who genuinely needs one, and how to decide without guessing.

• List what paid tiers typically add over free ones
• Judge whether your own usage justifies paying
• Avoid paying for several tools when one would do

A short decision tree that gets you to the right category in about five seconds, and an exercise that shows you the differences for yourself.

• Route a task to the right kind of tool without deliberating
• Recognise the tasks where the choice genuinely matters
• Compare tools on your own work rather than on someone else's benchmark

Accounts, apps, and the handful of habits that make the difference between using this occasionally and using it daily.

• Set up an account and the apps that suit how you work
• Find the six controls that exist in every assistant
• Adopt the habits that make the tool available at the moment you need it

Most disappointing AI output is a briefing problem, not a model problem. Four things to include, and the one habit that improves results more than any clever wording.

• Brief an AI tool the way you would brief a competent contractor
• Use the four ingredients that fix most weak output
• Iterate on a reply instead of starting again

Describing what you want gets you somewhere. Showing an example gets you there in one go — and it is the fastest way to stop output sounding like AI.

• Use an example of the output you want instead of describing it
• Set the voice by supplying a sample of your own writing
• Recognise when one example is worth more than a paragraph of instructions

Stop trying to write the perfect prompt. Five rounds of ordinary refinement beats one brilliant opening, and it is far less work.

• Refine output through iteration rather than restarting
• Ask the model what it needs before it answers
• Use self-critique to improve a draft without doing the editing yourself

Six failures you will meet repeatedly, what causes each, and the specific move that fixes it.

• Diagnose a bad result from its symptoms rather than guessing
• Apply the specific fix for each of the six common failures
• Recognise the failures that prompting cannot fix

Ten reusable shapes that cover most of what anyone ever needs. Learn these and you can stop collecting prompts.

• Recognise and apply ten reusable prompt patterns
• Adapt a pattern to your own work rather than copying a prompt
• Build a personal file of the ones you actually use

The single most useful thing these tools do for most people — and the four questions worth asking of any document you upload.

• Upload documents and ask questions that produce more than a summary
• Recognise where document handling degrades
• Use a summary as a map rather than as a substitute for reading

How to describe an image so you get what you pictured, and the four situations where using a generated image is a bad idea.

• Describe an image with enough specificity to get what you intended
• Judge when a generated image is appropriate and when it is not
• Understand the rights and likeness issues in plain terms

Transcription, meeting notes, and talking to a machine as a way of thinking — plus the consent question that comes with all of it.

• Turn a recording into notes, decisions and actions rather than a wall of text
• Use voice as a thinking mode rather than a typing substitute
• Handle consent before recording anybody

What changes when the tool can look things up, why citations still need checking, and what deep research modes are actually for.

• Tell whether a tool actually searched before answering
• Check a citation properly rather than trusting its presence
• Judge when a long research run is worth the wait

What a hallucination actually is, why it is a feature of how these systems work rather than a bug, and the six tells that give one away.

• Explain why invention is a consequence of how these systems work
• Recognise the six tells of a fabricated detail
• Identify the topics where invention is most likely

Checking everything is impractical and checking nothing is reckless. A tiered rule that takes seconds and covers both.

• Apply the stake-your-name test to decide how much checking is warranted
• Use a tiered rule matching effort to consequence
• Ask the model to separate what it knows from what it is inferring

Where the skew comes from, why it shows up as an unstated default rather than an obvious opinion, and how to surface it.

• Explain where bias in these systems originates
• Spot an unstated assumption in an otherwise reasonable answer
• Ask in a way that surfaces the default rather than accepting it

Why it can be confidently wrong about today, what a knowledge cutoff actually means, and when to force it to look.

• Explain what a knowledge cutoff is and what it affects
• Predict which questions will be stale
• Force a search when currency matters

Where your conversations go, the settings that actually matter, and the ten-minute audit worth doing before you paste anything real.

• Describe what typically happens to a conversation after you send it
• Find and set the four controls that matter
• Explain how consumer and business accounts usually differ

One rule of thumb, a short list of categories, and the eight scenarios people actually get wrong.

• Apply a single test to decide whether something should be pasted
• Recognise the categories that are never appropriate
• Redact well enough that the task still works

What is now cheap to fake, the specific scams that follow from it, and the one defence that works regardless of how good the fake is.

• Describe what can now be convincingly faked and how little is needed
• Recognise the markers common to AI-enabled scams
• Set up a verification habit that works even against a perfect fake

Complaint letters, disputed bills, confusing policies and the comparisons you have been putting off — the tasks where this pays for itself fastest.

• Draft a letter that gets a result rather than one that sounds cross
• Decode a bill, policy or contract you do not understand
• Compare options against criteria you actually care about

A patient tutor that never sighs, available at midnight — and the one way of using it that quietly stops you learning.

• Use explain-then-test loops rather than passive reading
• Read difficult material with an assistant alongside
• Recognise when it is helping you learn versus doing it for you

Genuinely useful for understanding and preparing. Not a substitute for a professional, and the line between those is sharper than it looks.

• Distinguish understanding a situation from getting advice about it
• Use it to prepare for a professional appointment
• Recognise the four conditions that make a question unsafe to rely on

Most people stop using these tools not because they do not work, but because they never decided what to use them for. Twenty minutes fixes that.

• Identify your own repeating tasks that suit these tools
• Choose three to hand over deliberately
• Recognise the tasks worth keeping for yourself

Custom instructions, memory and saved spaces — the step that stops you re-explaining yourself every single time.

• Write custom instructions that meaningfully change output
• Distinguish instructions, memory and project spaces
• Build a personal context pack you can reuse

How to follow a fast-moving subject without drowning in it — and the one project that turns this course into a habit.

• Follow changes without chasing every announcement
• Test new capabilities against your own benchmark
• Complete a real project end to end and reflect on it honestly