LLM as Study Partner

This concept changes how you study from this point forward. Instead of memorising facts on your own, you will learn to use a language model — the same kind of technology behind ChatGPT — as a personal tutor that explains things, questions you, and generates practice problems. This unlocks everything else in this course: statistics, hypothesis testing, and the mathematics of AI. The skill you build here — verifying what a model tells you against trusted sources — is exactly the skill you need for every later topic, because every later topic will invite you to ask the model for help. Master this now, and every future page becomes faster and deeper.

Imagine you are learning to cook a new dish. You have a recipe book, but the recipe assumes you know what 'julienne' means, and it skips steps. A good cook standing next to you would watch you chop, point out that your knife grip is wrong, ask you what you would do next, and give you a practice vegetable to try again. That cook is what an LLM study partner is like. You ask it to explain the step you do not understand, and it answers in plain words. You ask it to test you, and it gives you a question and checks your answer. It never gets tired, never judges, and it has read an enormous number of books. But like any cook who learned from books rather than from a kitchen, it sometimes tells you confidently how to do something that would ruin the dish. So the rule is: use the cook to practise and to clarify, but check the recipe book for anything that matters. The cook is a partner in your learning, not the authority on it. The point of having this partner is not to make learning easier by doing it for you — it is to make learning harder in a useful way, by forcing you to explain, to answer, and to practise, all with someone responding to each thing you say.

Think of an apprenticeship. In a traditional apprenticeship, a young craftsperson learns by working alongside a master. The master demonstrates a technique, then hands the apprentice the tool and says 'try it'. The apprentice tries, gets it wrong, and the master points at the exact moment of the mistake: 'You released the chisel there — that is why the cut jumped.' The apprentice tries again with the correction. This cycle — demonstrate, attempt, diagnose, correct — is the heart of learning. An LLM study partner runs the same cycle, but it plays the master imperfectly. It can demonstrate an idea in many different ways, which a human master sometimes cannot. It can generate practice problems by the hundred, each slightly different, which a human master could not do on the spot. And it can diagnose your answer, zooming in on the specific line of your reasoning where you went wrong. But where the analogy breaks is crucial: the human master has physical knowledge and has made mistakes themselves, so they know when you are about to cut yourself. The LLM has no body, no hands, and no memory of making mistakes. It cannot see you struggling, and it cannot know for certain whether what it is telling you is true. It will sometimes state a falsehood with complete confidence. So the apprentice in this relationship has a new duty that a traditional apprentice did not have: verify everything that matters against a trusted source. The LLM is the practice partner, not the master. The books are the master. This difference is not a flaw in the model — it is a property of how it works. It predicts the next word based on patterns in its training data, so it can generate a sentence that sounds true without having checked it against a fact database.

Using a language model as a study partner means treating it as an interactive tutor that explains concepts, asks you questions, and generates practice material, while you take responsibility for verifying everything it says against trusted sources.

You have not studied anything in this course yet, so this is your first concept. But you have almost certainly used a search engine, and you have almost certainly used a calculator. A search engine gives you a list of links; the study partner gives you a direct answer. A calculator computes a number; the study partner computes a sentence. The skill you build here — treating a generated text as a hypothesis to be checked, not as fact to be accepted — is the same skill you will use in 'Data Literacy' when you evaluate whether a statistic is trustworthy, and in 'Descriptive Statistics' when you ask the model to summarise a dataset and then check its arithmetic against your own. The library notes for this topic record three key points that this page is built around: explicitly building the model into teaching beats pretending students will not use it; verification is the skill being taught alongside the subject; and the interaction between you and the model surfaces misconceptions that traditional exercises miss. All three are reflected in how you will work.