Is Cosine-Similarity of Embeddings Really About Similarity?

Re:Is Cosine-Similarity of Embeddings Really About Similarity? by bongodongobob on Tuesday March 12, 2024 @03:35AM (#39677038)

Isn't that kind of the point though? That beautiful and ugly are encoded closely as an "idea" when viewed from the appropriate angle?

[ Parent | on HN ]

Re:Is Cosine-Similarity of Embeddings Really About Similarity? by Sai_ on Tuesday March 12, 2024 @03:43AM (#39677087)

Apart from an ESL class explaining antonyms, I can’t think of any use case where an API which could equally return “ugly” or “beautiful” can be used.

We need embeddings to give relatedness across axes like synonymity etc.

[ Parent | on HN ]