[vb7132]

Isn’t a cosine similarity over the text embeddings a much more effective way to handle categorization?

The whole reason why embeddings work so well is because they encode the underlying meaning of the texts