thanks, I'll try that. your comment prompted me to look further why averaging is effective, which led me to this article, which made the intuition more clear to

thanks, I'll try that. your comment prompted me to look further why averaging is effective, which led me to this article, which made the intuition more clear to me - "The mean is a reasonable way to summarize embeddings thanks to the blessing of dimensionality. Because an exponential number of embeddings are nearly orthogonal in high dimensions, it is unlikely for two independent collections to have similar averages. On the other hand, two related documents are guaranteed to have similar averages. As a result, two collections have similar averages if and only if they share many similar embeddings with very high probability." https://randorithms.com/2020/11/17/Adding-Embeddings.html#:~:text=Spoiler%20Warning%3A%20The%20average%20is,two%20random%20high%2Ddimensional%20vectors https://randorithms.com/2020/11/17/Adding-Embeddings.html#:~....