How much math do I actually need for machine learning jobs?

For applied roles, the honest bar: Must have (intuition level, not proofs): what a derivative/gradient means (why models 'descend' toward better weights), basic linear algebra (vectors, matrix multiply — enough to reason about embeddings and why GPUs help), probability basics (distributions, conditional probability, Bayes intuitively), and statistics you'll use weekly — mean/variance, sampling bias, significance, why train/test splits exist. Almost never used at work: deriving backprop by hand, eigenvalue proofs, measure theory. Interviews probe intuition: 'why does L2 regularization shrink weights?' wants a plain-words answer, not a derivation. Two months of 3Blue1Brown plus StatQuest covers the actual bar for most applied roles. Research roles are a different conversation entirely.