FOUNDATIONS • 27 / 397
Start before LLMs. Understand where AI, ML and neural networks fit.
Loss
Loss is a numerical measure of how wrong a model’s prediction is during training.
Think of it like
Think of Loss as one layer in a very large toolbox: you only need to understand what job that layer performs.
Real life
A phone spam filter, recommendation engine, fraud detector or voice assistant may use Loss somewhere inside.
SRE lens
Training tries to reduce loss across many examples.
Remember thisLoss is a numerical measure of how wrong a model’s prediction is during training.