What was the limitation of traditional search methods like keyword matching?	They relied on string similarity (e.g., synonyms or stemming) rather than semantic meaning.
What core concept do embeddings introduce to search, and how is meaning represented?	Meaning is represented as a place or set of coordinates in a multi-dimensional space.
How does an embedding model convert text into a usable format for search?	It converts the text (string) into a fixed-length list of numbers, called a vector.
In an embedding space, how is a search query executed to find relevant information?	By plotting the question and finding the nearest $k$ items (k-nearest neighbors) to that point.
Why are high dimensions (e.g., 1024) necessary in real-world embedding models?	To capture the complexity of many different topics and concepts beyond what a few axes can represent.
What is the nature of the axes (dimensions) in a real embedding model?	The meaning of the axes is emergent and unlabeled, meaning no human can assign a specific topic to a given dimension.
What is the primary mathematical measure used for similarity search in embedding space?	Cosine similarity.
What does Cosine Similarity measure, and how does it differ from straight-line distance?	It measures the similarity based on the *direction* the vectors point, not the physical distance between them.
What is the relationship between similar concepts and their coordinates in the embedding space?	Similar concepts will end up located near each other in the space.