Practice question from Semantic Search Principles
Explain why you cannot compare embeddings from two different models (e.g., BERT and OpenAI Ada-002) directly. What principle does this violate?
Answer
!AI
Explanation
Each embedding model creates its own unique vector space with specific dimensions and learned semantic relationships. Comparing vectors from different models is like comparing measurements in incompatible units—the numbers have no meaningful relationship. For valid similarity calculations, queries and documents must use the exact same embedding model.