Issue with text-embedding-004 Returning Identical Vectors for Specific Languages

[David_Lanzendorfer]

Embeddings are a vector representation of the semantics of expressions. So when you say good morning, it’s the same as おはようございます, so it makes sense that the vectors are basically the same, otherwise Euclidean proximity search wouldn’t work. It’s the whole point about positional encoding. If however, the Euclidean distance between Good Morning and Good Evening are zero, then something’s broken, because Good Morning and Good Evening differ in their semantics in the information what time of day it is.