Comment by amitness

Try this trick that I learned from Cohere: - Fetch top 10*k (i.e. 100) results using the hamming distance - Rerank by taking dot product between query embedding (full precision) and binary doc embeddings - Show top-10 results after re-ranking