I want to learn machine learning. Where do I actually start?

Start with a hands-on project, not a textbook. I've seen too many people stall out after three weeks of linear algebra videos, convinced they're not smart enough. The truth is, you don't need a PhD to be useful. You need to get your hands dirty. Pick a small, specific problem you care about. For example, try building a model that predicts whether a Titanic passenger survived based on their age, ticket class, and gender. This is the classic beginner project for a reason — the data is clean, small, and you'll learn the entire workflow: loading data, cleaning it, training a model, and seeing how well it did. You can do this in a free Google Colab notebook in an afternoon using a Python library called scikit-learn. A more modern route I've been recommending lately is fast.ai's free Practical Deep Learning course. It flips the script by having you build an image classifier in the first lesson, then gradually unpacking why it works. The insight here is that machine learning is more like carpentry than physics. You learn by building things, making mistakes, and gradually understanding the tools. Don't try to master Python first, then statistics, then calculus. You'll quit. Learn just enough Python to be dangerous, then immediately apply it. The math will make more sense once you've seen what it does. According to a 2025 Kaggle survey, over 60% of working data scientists are self-taught, which tells you the traditional academic path isn't the only way in.