That doesn't make you dumb; there is no intuitive meaning for the axes chosen; you can think of them, roughly, as statistically chosen to maximize clustering.
Statistically chosen to maximize *some particular loss measure, which in this case might be the t-SNE or UMAP criterion, and is computed only globally and not for different filters.
Right (I mean, I'm saying "right" but really I should just say "I'm taking your word for it"), but even more fundamentally this is dimensionality reduction from an OpenAI embedding vector, which seems almost like the asymptotic limit of inscrutability.
Comments
My dumb ass was trying to figure out what each dimension meant
That doesn't make you dumb; there is no intuitive meaning for the axes chosen; you can think of them, roughly, as statistically chosen to maximize clustering.
Statistically chosen to maximize *some particular loss measure, which in this case might be the t-SNE or UMAP criterion, and is computed only globally and not for different filters.
Right (I mean, I'm saying "right" but really I should just say "I'm taking your word for it"), but even more fundamentally this is dimensionality reduction from an OpenAI embedding vector, which seems almost like the asymptotic limit of inscrutability.
same