Five language models cluster mentions of the same entity and separate mentions of different entities with AUC 0.66-0.9, with entity identity most cleanly encoded in a low-dimensional linear subspace of early layers.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
On Entity Identification in Language Models
Five language models cluster mentions of the same entity and separate mentions of different entities with AUC 0.66-0.9, with entity identity most cleanly encoded in a low-dimensional linear subspace of early layers.