Fixed counterfactual explanation datasets train LMs such that generated explanations track the model's evolving behavior rather than the fixed targets, due to persistent correlation during training.
Are self-explanations from Large Language Models faithful?
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
years
2026 8verdicts
UNVERDICTED 8representative citing papers
Proposes SCSuff metric for evaluating LLM explanation sufficiency via model-generated alternative inputs, showing explanations are typically insufficient and predictable from hidden states.
Data-similarity and data-influence produce significantly overlapping rankings of training documents for LLM outputs, with asymmetry allowing a favorable cost-accuracy trade-off.
HANSEL extracts navigable evidence from agent trajectories with 83.7% precision and 88.8% recall on 45 tasks, reduces volume by 61.6%, and improves verification metrics in a 14-participant study.
iPOE generates and optimizes annotation guidelines from explanations to produce interpretable prompts, reporting up to 39% gains over baselines on four datasets with LLM explanations substituting for human ones.
VLMs show answer inertia in CoT reasoning and remain influenced by misleading textual cues even with sufficient visual evidence, making CoT an incomplete window into modality reliance.
Two linked user studies find that LLM rationale correctness and certainty framing affect trust and decision confidence while presentation format does not, and incorrect rationales increase gaze attention and pupil size.
LLMs show structured attribute-driven decisions that a behavioral model can predict, but self-reports recover those drivers only partially, indicating superficial beliefs.
citing papers explorer
-
Introspective Coupling: Self-Explanation Training Tracks Behavioral Change Despite Fixed Supervision
Fixed counterfactual explanation datasets train LMs such that generated explanations track the model's evolving behavior rather than the fixed targets, due to persistent correlation during training.
-
What LLMs explain is not what they believe: Evaluating explanation sufficiency under models' own input beliefs
Proposes SCSuff metric for evaluating LLM explanation sufficiency via model-generated alternative inputs, showing explanations are typically insufficient and predictable from hidden states.
-
Quantifying the Agreement Between Data-Influence and Data-Similarity to Understand LLM Behavior
Data-similarity and data-influence produce significantly overlapping rankings of training documents for LLM outputs, with asymmetry allowing a favorable cost-accuracy trade-off.
-
HANSEL: Extracting Breadcrumbs from Web Agent Trajectories for Interactive Verification
HANSEL extracts navigable evidence from agent trajectories with 83.7% precision and 88.8% recall on 45 tasks, reduces volume by 61.6%, and improves verification metrics in a 14-participant study.
-
iPOE: Interpretable Prompt Optimization via Explanations
iPOE generates and optimizes annotation guidelines from explanations to produce interpretable prompts, reporting up to 39% gains over baselines on four datasets with LLM explanations substituting for human ones.
-
Reasoning Dynamics and the Limits of Monitoring Modality Reliance in Vision-Language Models
VLMs show answer inertia in CoT reasoning and remain influenced by misleading textual cues even with sufficient visual evidence, making CoT an incomplete window into modality reliance.
-
When LLM Rationales Become User-Facing: Effects on Trust Perception, Decision-Making, and Gaze Behaviors
Two linked user studies find that LLM rationale correctness and certainty framing affect trust and decision confidence while presentation format does not, and incorrect rationales increase gaze attention and pupil size.
-
Superficial Beliefs in LLM Decision-Making
LLMs show structured attribute-driven decisions that a behavioral model can predict, but self-reports recover those drivers only partially, indicating superficial beliefs.