Shapley value analysis identifies powerful adjectives that steer MMLU performance in model-family-specific patterns, with non-additive interactions emerging in larger models.
Concept-level explainability for auditing and steering LLM responses
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Investigating Linguistic Steering: An Analysis of Adjectival Effects Across Large Language Model Architectures
Shapley value analysis identifies powerful adjectives that steer MMLU performance in model-family-specific patterns, with non-additive interactions emerging in larger models.