Persona axes derived from contrastive prompts and PCA yield linear probes that generalize better than raw-activation probes across 10 datasets for deception and sycophancy.
Findings of the Association for Computational Linguistics: EMNLP 2022 , address =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Do Linear Probes Generalize Better in Persona Coordinates?
Persona axes derived from contrastive prompts and PCA yield linear probes that generalize better than raw-activation probes across 10 datasets for deception and sycophancy.