Learn AI,
one lesson at a time.

You have a large language model and it is not doing what you need. The output is in the wrong tone, the format keeps drifting, or the model simply does not know enough about your domain to be useful. The temptation, especially after reading vendor documentation, is to reach immediately for fine-tuning: feed the model your data, retrain it on your use case, and get reliable results. That instinct is often wrong, and acting on it prematurely is an expensive mistake. Fine-tuning and prompting are not interchangeable tools that sit on the same shelf. They operate at fundamentally different levels of the model's architecture, solve different categories of problem, and carry very different costs, so choosing between them is one of the most consequential technical decisions a practitioner makes. As of 2026, frontier instruction-tuned models respond well to structured, detailed prompts, and many tasks that required fine-tuning in earlier generations are now solvable with prompting alone. At the same time, there are situations where no amount of careful prompting will get you where you need to go, and recognising those early saves significant time and money. This lesson gives you a clear mental model for making that call.

A teacher sits with an essay that does not sound like the student who submitted it. A detection tool reports that the text is 92 per cent likely to be AI-generated. What happens next matters enormously, for that student and for the integrity of assessment itself, and the honest answer is that the detection score should play almost no part in it. Generative AI has not created academic dishonesty, but it has broken the assumption that most assessment quietly relied on: that a polished piece of written work is reliable evidence of the thinking that produced it. This lesson takes the problem in three steps. First, why AI detection tools cannot carry the weight that institutions are tempted to place on them. Second, why clear, task-level expectations do more for integrity than suspicion ever will. Third, how to redesign assessment so that it keeps measuring what it was always meant to measure: what the student can actually do.

You have already cleared up one of the biggest confusions in AI: that it is not one thing, but a family of related technologies, each with its own methods and limits. Now comes the question most people do not know they can ask: how does any of this actually work? Not in a hand-wavy sense of 'it learns from data', but in a real sense. What is the model doing when it learns? What happens inside the system when you type a message and press send? Why does it sometimes produce confident, fluent nonsense, and why can it not remember your conversation from yesterday? These are not trick questions. They have clear, non-mathematical answers, and understanding them changes how you use AI tools, how you evaluate claims about them, and how you think about their limits. Here is the central insight this lesson builds towards: modern AI models are not databases that look things up, not rule-following systems executing step-by-step instructions, and not reasoning engines in any philosophical sense. They are pattern-extraction machines, trained to recognise statistical regularities in enormous quantities of data and to use those patterns to generate plausible-seeming outputs. That might sound deflationary, but it is the opposite. The fact that this approach produces systems that can write code, translate languages, analyse images, and hold coherent conversations is genuinely remarkable. This lesson covers the full chain: from the basic logic of machine learning, through the structure of neural networks, to the architectural innovation that made modern large language models possible, and finally to what actually happens when you use one. By the end, you will have a mental model that holds up under scrutiny.