GrAInS uses Integrated Gradients to identify the most influential tokens, then builds layer-wise steering vectors that improve truthfulness, reduce hallucination, and preserve general capabilities in LLMs and VLMs.
A diagnostic study of explainability techniques for text classification
1 Pith paper cite this work, alongside 164 external citations. Polarity classification is still indexing.
1
Pith paper citing it
164
external citations · OpenAlex
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs
GrAInS uses Integrated Gradients to identify the most influential tokens, then builds layer-wise steering vectors that improve truthfulness, reduce hallucination, and preserve general capabilities in LLMs and VLMs.