You now know what a language model is. This lesson is about using one well: picking an assistant, understanding the controls it gives you, and learning what it is good for in a normal working day. Everything later in the course builds on this.
An assistant is a model plus a product
When people say "ChatGPT", "Claude" or "Gemini" they mean two things at once:
- The model: the trained network that predicts text.
- The product around it: the chat window, file uploads, web search, saved conversations, memory, and the tools it can use.
Two products can use similar models and still feel very different, because of the product features. A model can also change behind the same product name when the provider updates it. So treat any single review of "the best AI" as a snapshot, and expect to re-check every few months.
The choices that matter
| Choice | What it means | When it matters |
|---|---|---|
| Fast versus strong model | Many assistants offer a quick everyday model and a stronger, slower one | Use the strong one for analysis, long documents, tricky code and anything you will rely on |
| Reasoning or "thinking" mode | The model works through a problem before answering | Maths, planning, debugging and multi-step questions |
| Web search | The assistant can look things up and cite pages | Anything recent, and any fact you may need to check |
| File and image upload | You can attach documents, spreadsheets, screenshots | Summaries, extraction, "what is wrong in this screenshot" |
| Projects or custom instructions | Standing instructions and files kept across chats | Repeated work such as a weekly report |
| Memory | The assistant remembers facts about you between chats | Convenient, but a privacy decision (lesson 6) |
| Voice, camera, screen sharing | You talk or show instead of typing | Quick questions and live troubleshooting |
Names and limits change often, and free plans are usually more limited than paid ones. Check what your plan includes before you decide something is "not possible".
Which one should you pick?
For learning, the three most widely used families are Anthropic's Claude, OpenAI's ChatGPT and Google's Gemini. Microsoft Copilot is common inside Microsoft 365 workplaces, and many organizations provide an approved internal assistant.
A practical approach:
- Start with what your organization allows. If your company has an approved assistant, use that for anything involving work information.
- Use a free plan of one or two others to learn. You will quickly find which style you prefer.
- Choose one main assistant and get good with it, rather than collecting subscriptions.
- Compare on your own tasks. Give the same real prompt to two assistants and judge the answers yourself. Your tasks tell you more than a leaderboard.
Do not paste confidential, customer or regulated information into a personal account of a public assistant. Use the tool your organization approves. You will build a clear policy for this in lesson 6.
What assistants are good at
Strong fits, because the answer can be checked quickly and a first draft saves time:
- Drafting and rewriting: emails, documentation, release notes, summaries for different audiences.
- Explaining: a concept, an error message, an unfamiliar log format, a contract clause in plain words (then check with an expert).
- Summarizing and extracting: key points from a long document, action items from meeting notes, fields from messy text.
- Brainstorming: options, test cases, names, risks you may have missed.
- Transforming: reformatting data, converting a table to JSON, translating tone or language.
- Coding help: scripts, regular expressions, SQL, explaining code, writing tests.
What to be careful with
- Facts you cannot check, especially numbers, quotes, citations and legal or medical claims.
- Anything recent unless web search is on and you read the sources.
- Exact arithmetic and counting unless the assistant runs code for it.
- Decisions with serious consequences. Use the assistant to prepare, not to decide.
Try it: a two-assistant test
Pick a real task from this week, such as "summarize this changelog for non-technical staff". Write one clear prompt. Run it in two assistants, or in the fast and strong model of one. Note which answer you would send with the fewest edits, and why. Keep this note: it starts your playbook in the capstone.
Go deeper (optional)
A long, practical walkthrough of how an expert uses assistants day to day: model choice, thinking models, search, file upload and more. Optional, but worth watching in parts. The specific products will have changed, the habits have not.
Resources
A plain guide to choosing an assistant, switching to the stronger model for serious work, and using features such as deep research and voice. Model names date quickly, the advice keeps its value.
Free learning material on working with Claude. Browse the catalogue for what suits your role.
A free course on collaborating with AI effectively and responsibly. Its four habits, delegation, description, discernment and diligence, come back in later lessons.
A short lesson with exercises on the basics of prompting.
Check yourself
- What is the difference between a model and a product built on it? Give two features that come from the product.
- List three tasks from your own week that suit an assistant, and one that does not. Say why.
- Run the two-assistant test on one real task. Which answer won, and what made it better?