How much maths do you really need?

The amount of maths you need depends entirely on the role you are aiming for. An AI/GenAI engineer who builds products on top of LLM APIs needs very little maths. A research scientist pushing the state of the art needs strong linear algebra, calculus and probability. Most people reading this are closer to the first camp.

Do not let maths anxiety stop you from starting. But also do not believe anyone who says maths does not matter at all — once you go beyond using APIs, understanding the maths helps you debug, optimise and make better design choices.

Here is an honest breakdown of what each role actually requires in practice.

Three areas cover 90% of what working ML engineers need. Linear algebra (vectors, matrices, dot products) — because all data and model parameters are stored as matrices. Probability and statistics (distributions, Bayes theorem, hypothesis testing) — because ML is fundamentally about making predictions under uncertainty. Basic calculus (derivatives, gradients, chain rule) — because training neural networks uses gradient descent.

You do not need to derive proofs. You need to understand what a gradient is, why a loss function decreases during training, and what it means when someone says 'the model is overfitting'.

Unless you are going into research: differential equations, real analysis, topology, abstract algebra, and most of measure theory. These are beautiful branches of mathematics but they are not required for applied AI engineering work.

Learn maths in context, not in isolation. Do not take a full linear algebra course before touching ML — instead, learn the linear algebra concepts as they come up in ML tutorials. 3Blue1Brown's Essence of Linear Algebra series is the best visual introduction. Khan Academy covers probability well. Then apply what you learn immediately in code using NumPy.