1 paper · 2 filters
Isabel Papadimitriou, Jacob Prince
Large language models (LLMs) work by manipulating the geometry of input embedding vectors over multiple layers. Here, we ask: how are the input vocabulary representations of langua…