they don't have meaning by themselves. they are two dimensions that umap projected the original embeddings down to in order to show a combination of local neighborhood similarity or closenes
Well, they do have meaning by themselves, but it's more work to figure that out. All regular, predictable relationships "have" meaning because all meaning is prescribed. And since we've captured many such prescriptions in LLMs, they can do a decent job approximating those.
Comments
they don't have meaning by themselves. they are two dimensions that umap projected the original embeddings down to in order to show a combination of local neighborhood similarity or closenes
Well, they do have meaning by themselves, but it's more work to figure that out. All regular, predictable relationships "have" meaning because all meaning is prescribed. And since we've captured many such prescriptions in LLMs, they can do a decent job approximating those.