The dialogue between model theory and machine learning theory has a rich history dating back to the seminal work of Laskowski (1992) in the binary classification setting. Ben Yaacov (2008) transported this flavour of correspondence into continuous logic; notably, unlike in the discrete case, the natural analogue of NIP in the continuous setting corresponds to a dimension which does not characterize learnability. We will see why this is the case from the machine learning perspective, and we will introduce several other useful notions of dimension in this setting towards achieving a modern bird’s-eye-view of the field.