Comment by yobbo

True, on a hypersphere cosine and euclidean are equivalent.

But if random embeddings are gaussian, they are distributed on a "cloud" around the hypersphere, so they are not equal.