park2025-final-layer-final-token
IN premise — summaries/2026/08/24/park-2024-categorical-hierarchical-concepts-s1-we-show-how-to-move-from-representations-of-binary-concepts-.md
Created 2026-08-25T02:58:24+00:00
LLM representations for concept geometry in Park et al. (2025) are read at the final layer, final token position of the transformer.
Summary
Park et al.'s geometric analysis of how concepts are organized inside an LLM is pinned to one specific spot in the model: the output at the very last layer and the very last token position. Every downstream claim about the shape, distances, and relationships between concept representations depends on that single measurement point, so changing where you read would change the entire geometric picture.