Activation steering guided by verbal-versus-symbolic CoT decomposition improves math accuracy by a few points, but the SAE-free derivation conflates L1 and L2 objectives.
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
REJECT 1representative citing papers
citing papers explorer
-
Feature Extraction and Steering for Enhanced Chain-of-Thought Reasoning in Language Models
Activation steering guided by verbal-versus-symbolic CoT decomposition improves math accuracy by a few points, but the SAE-free derivation conflates L1 and L2 objectives.