Tag
This paper reveals that the optimal layer for linear probing to read concepts differs from the optimal layer for activation steering in omni-modal large language models, challenging common heuristics in representation engineering.
This paper investigates internal representational differences between RL and SFT fine-tuned models on mathematical reasoning, finding that RL models exhibit more linearly separable hidden states and hierarchical layer importance. Token allocation variability under repeated sampling suggests training pipeline dependence rather than RL vs SFT alone.
This paper presents a method that uses frozen medical large language model (LLM) representations as a shared embedding space to predict primary ICD diagnosis categories from both structured and unstructured electronic health record data, achieving improved accuracy over baseline methods on MIMIC-IV and showing transferability to MIMIC-III.
This paper introduces a dual diagnostic framework to trace the internal lifecycle of code reasoning in LLMs, revealing that models first 'brew' answers and then diverge into four resolution outcomes, with stable brewing across architectures but varying resolution success.
This paper introduces 'fragility', a complementary metric to probe accuracy that measures activation-noise level at which probe accuracy collapses, enabling analysis of representation evolution during LLM pre-training even after accuracy saturates.
This paper demonstrates that linear probes on LLM hidden states detect task format confounds (e.g., source identity, response length) rather than distinct reasoning modes, using residualization and causal steering to show that high probe accuracy is due to superficial features, not computational structure.