Adding Dimensions, On Purpose
Take a cat. a_cat = np.random.rand(64, 64, 3) gives you a 64×64 RGB image — a respectable 3-D representation. Good cat. Three dimensions, ~12,000 numbers, captures the visual surface.
Now put glasses on the cat. The glasses are also data — frame color, lens tint, position offset, ridiculousness factor. Adding glasses isn't replacing the cat. It's composing a new entity by merging the cat's dimensions with the glasses's dimensions. cat_with_glasses = combine(cat, glasses).
Add a hat. Now cat_with_glasses_and_hat. Push further: a personality, a mood, a name. The dimensions stack. The cat-ness gets richer.
Push enough and you arrive at a Cat-Woman — a high-dimensional being whose feline-ness, human-ness, theatricality, and backstory all coexist as merged dimensions. You didn't add a new species. You composed an instance from dimensions you already had.
Tangible × Abstract
The leap that breaks people: cuteness is also a dimension. So is vibe, cringe, fashion sense, main character energy. These aren't physically located on the cat the way fur color is — but in a model that has to understand the cat, they're as real as pixels.
Modern AI takes this idea to thousands of dimensions per object. A word in GPT-class models lives in ~4,096-dim space. An image in vision-language models lives in similar territory. Most of those axes don't have human-readable names — the model invented them while learning. They're dimensions of meaning, learned from data.
The Object-Oriented Mold
Notice we're already doing OOP without saying so:
Catinherits fromMammalinherits fromAnimal.CatWithGlassescomposesCat+Glasses.CatWomanmulti-inherits fromCatandWoman(welcome to multi-mol mold-stacking).
The same instinct — compose richer instances from cleaner molds — drives both Python class design and AI feature engineering. They're dialects of the same idea.
class Mr.Nobody():
class Mr.NobodyAge10(Mr.Nobody):