3 ms·
Ooh that's an interesting idea. Another way to do this could be to calculate the "average" of all of the embeddings of documents in the index, then try to figu
by simonw 2y ago
Ooh that's an interesting idea.
Another way to do this could be to calculate the "average" of all of the embeddings of documents in the index, then try to figure out if the query embeds to a point that's a significant distance from that average. Maybe figure out that distance through experimentation?