Comment on Understanding Pgvector's HNSW Index Storage in PostgresparentComments−yawnxyz2yyeah usually much better results getting a list of ~10 fuzzy keyword search results, 10 semantic/embeddings results, and using something like Cohere rerank (or just a cheap GPT model) to choose the best 5-10 results from the pile
Comments
yeah usually much better results getting a list of ~10 fuzzy keyword search results, 10 semantic/embeddings results, and using something like Cohere rerank (or just a cheap GPT model) to choose the best 5-10 results from the pile