Skip to content
AI LAB
3/5
LAB 03 / EMBEDDINGS

Meaning becomes geometry.

Pick a word below. Related words cluster nearby, unrelated ones sit far apart.

This lab uses a small, hand-placed 2D space so the geometry is easy to see, not output from a real embedding model — production embeddings use hundreds or thousands of dimensions. The same idea works for whole sentences, documents, images and audio too, not just single words.

Embedding
A vector (list of numbers) representing characteristics a model learned from data. Text with related meaning often ends up with vectors that point in a similar direction, according to a similarity measure. Individual numbers in the vector don't usually mean anything on their own, it's the whole vector's direction that matters.
Vector space
The geometric space those vectors live in. Every piece of text becomes a point; direction and distance between points often reflect how related their meanings are, though the exact relationship depends on how the embedding model was trained.
Cosine similarity
Measures the angle between two vectors, from -100% (pointing in exactly opposite directions) through 0% (perpendicular, unrelated) to 100% (pointing in the same direction). It's a geometric measurement, not a meaning judgement — but for a well-trained embedding model, a higher score usually does correspond to more closely related meaning.
dog

Click a word to focus it. Click a second to compare, or click the same word again to clear the comparison.

Pick a second word above to compare it against “dog”.

Nearest to “dog”

Next04 RAGHow do we give it knowledge it wasn't trained on?