> Why is m a string/key matching function? Why can’t we use more well-defined math function, e.g. Euclidean distance, cosine function?... Isn't the purpose of

> Why is m a string/key matching function? Why can’t we use more well-defined math function, e.g. Euclidean distance, cosine function?... Isn't the purpose of a search index (aka inverted index) to compute the cosine similarity efficiently? Is this not possible to do for latent space dense vectors? Or am I missing something?