Reaction
Doubts about using 256 features to pre-index search data
A commenter says an approach needing fresh inference on all search data at search time wouldn’t qualify.
TLDR
One commenter equates interpretable embeddings with “jev.” In a reply, another says their previous work on classifier-based embeddings isn’t quite the same. They doubt 256 features are enough to pre-index anything, though they say well-chosen features could prove them wrong.
Combined views
—
1 Source, first seen ago
