[physicsguy2]

Completely agree! It feels like a gimmick to make things sound fancier than they are.

A key property of (actual) tensors is how they transform under coordinate transformations. One can store the components in multi-dimensional arrays but that doesn't fully describe the tensor.

Vectors which are a kind of tensor (of type (1,0) i.e. maps from the dual space to the field) also transform in a very certain way under coordinate (change of basis) transformations.

In physics, every quantity has certain transformation properties under rotations, Lorentz transformations, internal symmetry transformations and that puts a tight constraint on the quantities that can be constructed and the "tensorness" or "vectorness" of quantities has deep meaning. The use in deep learning completely bypasses all the meaning and instead defines a tensor to be a high-dimensional matrix.