A vector is an ordered list of numbers. When an embedding model processes text, it outputs a vector — a fixed-length list of floating-point numbers. For example, the sentence "The cat sat on the mat" might become a list of 1536 numbers like [0.12, -0.45, 0.87, ...].
Each number in the list is called a dimension. A 1536-dimensional vector has 1536 numbers. The vector as a whole represents the semantic content of the input.
Individual dimensions do not have simple human-interpretable meanings. You cannot look at dimension 47 and say "that's the happiness dimension." Meaning emerges from the pattern of all dimensions together.
During training, the model learns to distribute meaning across dimensions so that similar content produces similar vectors. Semantic relationships in language correspond to geometric relationships in the vector space — that's the core insight.
Human intuition breaks down in high dimensions — we can visualize 2D and 3D easily, but a 1536-dimensional space is impossible to picture. Mathematically, all the same rules apply: distance, direction, and angle all work the same way with more coordinates.
High dimensionality is actually a feature. More dimensions give the model more capacity to encode subtle semantic distinctions. A 1536-dimensional space can represent far more nuanced relationships than a 64-dimensional one.
Most values are zero. A 100,000-word vocabulary produces a 100,000-dimensional vector where nearly all values are 0 except the words present in the document.
Simple to compute. Does not capture meaning — only word presence.
Nearly all dimensions have non-zero values. A 1536-dimensional embedding has 1536 meaningful numbers.
Encodes continuous, nuanced meaning. Semantically similar content produces geometrically similar vectors.
Ask the AI assistant about vectors, dimensions, or how high-dimensional spaces work.