Fifteen minutes of setup. Ten minutes of reading.

Checklist

1. Laptop and charger. A tablet works for most exercises; a laptop is easier for the data exercise. Your phone is handy for the live votes, which run in ClassPulse: scan a code, no app or account.
2. Log in to at least one AI tool you’re allowed to use for work. Microsoft Copilot, ChatGPT, Claude or Gemini are all fine. Two is better, because the first exercise compares them. Free plans work for most exercises, though some limit file uploads; check yours before the day. You won’t need to enter anything confidential. No approved tool? On the day you can use the class chat: a local model and Claude Sonnet side by side, open for registration on course days only. Chats are stored on the course server and deleted after the course, so don’t paste anything confidential.
3. Check you can upload a file to that tool, or paste a few hundred lines of text into it.
4. Have a spreadsheet app (Excel, Google Sheets or Numbers). You’ll use it to check the AI’s numbers.
5. Think of three tasks you do every week that take 30 minutes or more. In the afternoon you’ll plan how to deliver one of them with AI.
6. Bring one decision you make over and over, and three recent times you made it. Something small and routine: which enquiries get a same-day reply, which invoices you query, which requests go to a senior colleague. For each of the three, jot down what you knew at the time and what you decided. Leave out names and client details. In Exercise 6 you’ll write down your rules and test them against these cases. If none of your work can go into an AI tool, use one of the decision cards instead.
7. Find out what your organisation allows. Which AI tools are approved, and what kinds of information can go into them?
8. Optional: take the two-minute Where’s your edge? self-check, and play the trust tool game once at trusttool.borck.education (ten scenarios, about ten minutes).

AI in ten minutes

AI is a bigger pie

“AI” covers a lot. Machine learning learns patterns from data rather than following hand-written rules; it’s behind fraud scores, demand forecasts, spam filters and recommendations, and it’s probably already running inside your organisation. Deep learning uses many-layered neural networks for speech, images and translation. Generative AI produces new text, images, audio or code, and large language models are generative AI for language. Natural language processing and computer vision are fields that cut across all of these.

The day focuses on large language models (ChatGPT, Claude, Gemini, Copilot) because you can try them yourself on a laptop, in plain language. The judgement you’ll practise applies to any AI output, though: a fraud score, a forecast and a chatbot reply all need to be placed and checked.

What these tools actually do

Chat assistants such as ChatGPT, Claude, Gemini and Copilot are built on large language models. A language model is trained on an enormous amount of text and learns to predict which words are likely to come next. It is then tuned to follow instructions and be helpful. Ask it a question and it produces the most plausible response to that question, one word at a time.

Modern assistants add more on top. Many can search the web, read your files, and write and run small programs to do calculations. That has made them much better at arithmetic and current events than they used to be. Underneath, though, the core is still prediction, and that explains both what they’re good at and how they fail.

What you get back is a convincing average: fluent, well-structured and usually about right. For a first draft, a summary or a list of ideas, about right is often exactly what you need. For a contract figure, a citation or a client’s account balance, about right is the same as wrong.

Where they help, and where they go wrong

The trap: mistakes that don’t look like mistakes

You’ve spent your career learning to spot weak work. A rushed draft looks rushed; gaps show. AI’s wrong answers don’t look like that. They arrive polished, complete and confident, with the same tone as its right answers. So the instinct you normally rely on stays quiet exactly when you need it.

The fix is to decide, quickly and deliberately, how much checking each output needs, and to do that check from outside the tool. Asking the AI “are you sure?” doesn’t count, because it tends to agree with itself. That’s what the first session of the day is about.

The tools you’ll meet

• Chat assistants you visit: ChatGPT, Claude, Gemini, Microsoft Copilot.
• AI inside software you already use: Copilot in Word, Excel, Outlook and Teams; Gemini in Google Docs, Sheets and Gmail.
• Agents: AI that works in a loop with tools, planning, acting and checking until a job is done. They are now built into some workplace software.

You don’t need to master a particular product. The skills in this course (deciding how much to trust an output, briefing well, checking from outside, choosing where AI fits) work on all of them, including the ones that ship next.

Data and privacy

• Use what your organisation allows. Business and enterprise plans usually keep your data out of model training. Many personal plans use your chats for training unless you turn it off. Settings change, so check the data controls of whatever you use.
• Classify before you paste. Public, internal, confidential or regulated? The real risk is where the data goes and who keeps it, the same as with email.
• Leave out what identifies people or clients. If you can’t share the data, describe its shape instead.
• Assistants remember. Many keep saved memories and inferred preferences, separately from your chat history. Deleting a chat doesn’t necessarily delete what was learned from it. A temporary or private chat avoids memory altogether.
• Keep work and personal apart. The Responsible use page has a short setup checklist: two accounts, training off, memory off for sensitive work.

Eight words you’ll hear on the day