Foundations: how a model actually learns
Every model is fitted to data, and the whole craft is in how closely. Fit too loosely and it misses the pattern; fit too tightly and it memorises the noise in your particular sample and fails on anything new. Drag the complexity control and watch both errors move: training error falls the whole way, while error on unseen data falls and then climbs again. The gap between those two curves is overfitting, made visible. Learn to read that gap and you can diagnose most model failures before touching a line of code — it is the same picture behind every “it worked in testing and broke in production”.
AI Foundations
How a model learns: under- vs over-fitting
A model that's too simple misses the pattern; one that's too complex memorizes the noise. Add data, add noise, and dial complexity to feel the trade-off that sits underneath all of machine learning.
More examples make over-fitting harder.
Real data is messy — noise is what a complex model wrongly memorizes.
Polynomial degree. Low = stiff line; high = wild wiggles.
With enough data and the right complexity, the fit tracks the true curve and generalizes.