Working With LLMs · for people who already use AI at work
AI isn't magic · it's leverage. Learn to operate it like an expert.
A hands-on course in using AI at work with good judgement: which tasks to hand over, how to check what comes back, and where these models fail, so their mistakes do not become yours.
- Seven evenings of about two hours
- No coding
- Works on your phone
- $30.00, one payment
Lesson 1.1 · predict first
We asked an AI model how many times the letter a appears in "Yokohama". What did it say?
- 2
- 3
- 4
- It refused
model · 5 recorded runs
answer: 3 · 3 · 3 · 3 · 3
actual: 2 (y-o-k-o-h-a-m-a)
The model never sees letters, only chunks of text called tokens. Knowing why it fails tells you where to check its work.
Make your own guess- Control the context. Command the output.
- From prompt guesswork to proven workflows.
- Catch AI's mistakes before your boss does.
- Turn AI hype into measurable ROI.
- Stop reviewing junk AI drafts.
- Bad prompts burn manager's time.
- Delegate to AI. Verify with rigor.
- Quality code starts with good input.
- Don't let AI write your reputation.
- AI assists. You are accountable.
- Smarter prompts mean shorter days.
- Work with AI, not around it.
- Output is only as good as context.
- Great outputs need great inputs.
- Verification is the new skill.
- Drive the LLM. Don't ride shotgun.
- Garbage prompts, garbage code.
- Don't ask nicely. Command it.
- Stop wishing. Start engineering.
- LLMs don't care about your job.
- AI wrote it. You own the blame.
- Copypaste is not a skill set.
- It's confidently wrong. Are you?
- Hallucinations aren't a strategy.
- It's a language model, not a deity.
- Prompting isn't a personality trait.
- The AI didn't fail. Your prompt did.
- Fancy words, zero actual reasoning.
- You gave it garbage. Surprised?
- Don't CC the team on hallucinated stats.
- AI confidentially ruining your career.
- It sounds smart until you read it.
- ChatGPT won't attend your disciplinary.
- Great code! Too bad it doesn't run.
- Probability engine, not a brain.
- Context windows don't grant wisdom.
- Auto-complete on high dosage.
- It's making things up. Read it.
- Your manager can tell you used AI.
- Confident is not the same as correct.
- Hand over the work, not your judgement.
- Stop prompting. Start directing.
- Off the record means off the prompt.
- LLMs don't fact-check their own lies.
- AI can't interview primary sources.
- Byline by you, liability by you.
- Deepfakes love sloppy reporting.
- Unverified claims break newsroom trust.
- Slapping copy-paste on press releases?
- ChatGPT isn't a reliable anonymous source.
- Ethics over speed. Verify every line.
- Every AI answer, checked like a pro.
- From AI user to AI professional, in a week.
- Brief it. Check it. Own it.
The habit the course builds
Know which answers to check, and how.
The trust ladder ranks what a model gives you by how often it goes wrong. You learn it in Module 5 and use it in every workflow after.
- 1Rewording or summarising text you suppliedHigh trust
- 2Explaining a well-established conceptMedium-high
- 3Facts, numbers, dates, rules and dosesLow
- 4Citations, quotations and linksVery low
- 5Recent events, or anything about youNear zero
Try it now
This is the opening of the first lesson. Make a guess first, then see what a model actually said, and why.
Predict first
We asked an AI model how many times the letter a appears in "Yokohama". What did it say?
Committing first is the point: a guess you wrote down is a guess you can learn from.
What you will be able to do
- 1Explain to a friend how an AI model actually produces a sentence, and predict where it will fail before it does.
- 2Turn a vague request into a brief that reliably gets what you meant, and work out why a prompt failed.
- 3Pick the right technique for the task (examples, step-by-step reasoning, structured output, breaking it down) and know when each is wasted effort.
- 4Say what retrieval, tools and agents add to a bare model, and which one fixes your problem.
- 5Check AI output against sources, catch fabrication, and measure how often the model is right on your own task.
- 6Decide, and defend, when to use AI and when not to, with integrity, privacy and bias in view.
- 7Design and document an AI-assisted workflow for a real task in your own field.
One week, two hours an evening
Every module is open from day one. Prefer slower? Switch to a 2, 4 or 6 week pace whenever you like.
- Day 1 · 104 minHow LLMs actually work
- Day 2 · 77 minPrompting fundamentals
- Day 3 · 82 minAdvanced techniques
- Day 4 · 94 minRetrieval, tools and agents
- Day 5 · 84 minVerification
- Day 6 · 121 minResponsible use, and your own workflow
- Day 7 · 105 minRun it, present it, finish
How it works
You learn more from watching a model fail than from watching it succeed.
Short lessons, one idea each
Most lessons take 8 to 14 minutes and start with a guess, not a lecture. No videos, so they load quickly on a phone and a slow connection.
A built-in AI playground
Run real models inside the lessons. Nothing to install and no separate AI subscription. Each exercise shows the runs you have left, and recorded runs take over if you use them up.
Your own pace
Every module is open from day one. The default plan is one week at about two hours a day, with 2, 4 and 6 week paces if life gets in the way. Change it whenever you like, and pick up exactly where you stopped.
A failure in every module
Each module includes a deliberate failure you predict, watch and diagnose, so you learn where these models break before it costs you.
What you leave with
Working people and students in any field who already use AI tools, or are about to.
A portfolio of your own work
A failure log, a prompt brief, an audit of AI output, a small evaluation and a workflow for your own field, each with an honest note of how you used AI.
A mastery map
Each of the seven outcomes shown as practised, demonstrated or applied, plus a before-and-after comparison of where you started.
A certificate of completion
It states exactly how you were assessed: automated checks and AI-generated feedback, with no human grading.
One payment
$30.00
- Every module and lab, including lessons added while your access lasts
- An AI allowance of about 70 playground runs, never shown in money
- Recorded runs for every exercise, so you can finish even if the allowance runs out
- Automated feedback on every question and labelled AI feedback on most written work
Access lasts 12 months from the day you enrol.
Enrol nowQuestions
Do I need to code?
No. Everything happens in your browser through plain-language prompts. There is no code to write anywhere in the core course.
Do I need my own ChatGPT or Claude account?
No. The course has a built-in playground that runs real models for you, paid for out of your course fee. You can practise in other tools too if you like, and the course shows you how to say so honestly.
What is the AI allowance, and what happens if I run out?
Each exercise has a small number of live runs, and the course as a whole allows about 70, with a daily limit. Nothing is shown in money. If you run out, or the playground is paused, the exercise switches to recorded runs, so you can always complete the course.
How long does it take?
About 11 hours in total. The default pace is one week at about two hours a day; you can also spread it over 2, 4 or 6 weeks.
Can I do it on a phone?
Yes. Every lesson is designed for a phone screen first, there are no videos, and pages are light enough for a slow connection.
Is there a teacher?
No. This is a self-paced course. Every question explains its answer automatically. Most written work gets AI-generated feedback that is clearly labelled as such and never decides your result on its own; the rest you check yourself against a checklist and a strong example.
What does the certificate say?
It is a certificate of completion for Working With LLMs, with your name and the date. It states how you were assessed: automated checks and AI-generated rubric feedback, with no human grading. It is not an academic credit or accreditation.
Which AI model does it use, and will the skills transfer?
The playground uses Claude models from Anthropic. The course itself is model-agnostic: what you learn about how these models work, how to brief them and how to check them applies to ChatGPT, Gemini, open models and whatever comes next.
How do I pay, and what is the refund policy?
Tap "Enrol now" and pay by card. Access opens as soon as the payment is confirmed. If the course is not for you, ask for a refund within 7 days of paying, as long as you have not yet passed Module 1's Prove it check.
What happens to what I type?
What you type into the playground is sent to Anthropic, which processes it in the United States to produce the reply. It is stored with your account so you can see your own history and so we can prevent misuse. It is never used to rank you against other learners. Do not paste other people's personal data or anything confidential; the course explains why.