Vector Space
In mathematics, a set composed of vectors; in natural language processing, it denotes the semantic space used to represent words or texts. Words are mapped to high-dimensional vectors, and semantic similarity is measured by the distance or direction between vectors.
Timeline (last 90 days)
Inside an LLM, the input word sequence is converted into a sequence of high-dimensional vectors, and the relationships between words are encoded by the geometric relationships of the vectors (distance, direction, projection).
When given extremely abstract prompts that lack specific constraints, generative models tend to output the highest-probability, most representative result—the centroid of that concept in vector space.
In vector space, each word, phrase, and abstract concept is mapped to a vector in high-dimensional space, typically ranging from hundreds to thousands of dimensions.
The unreadability of vectors makes dirty data difficult to detect intuitively; a 1536-dimensional floating-point array is meaningless to humans, and relationships between data points can only be measured through mathematical metrics such as cosine similarity and Euclidean distance.
All Facts (4)
Inside an LLM, the input word sequence is converted into a sequence of high-dimensional vectors, and the relationships between words are encoded by the geometric relationships of the vectors (distance, direction, projection).
50%UnverifiedWhen given extremely abstract prompts that lack specific constraints, generative models tend to output the highest-probability, most representative result—the centroid of that concept in vector space.
50%UnverifiedIn vector space, each word, phrase, and abstract concept is mapped to a vector in high-dimensional space, typically ranging from hundreds to thousands of dimensions.
50%UnverifiedThe unreadability of vectors makes dirty data difficult to detect intuitively; a 1536-dimensional floating-point array is meaningless to humans, and relationships between data points can only be measured through mathematical metrics such as cosine similarity and Euclidean distance.
50%