Introduces BonaFide benchmark of 3,066 ground-truth labeled CoTs showing most faithfulness metrics perform near chance with biases and poor scaling to longer chains.
Faithfulness tests for natural language explanations
4 Pith papers cite this work, alongside 17 external citations. Polarity classification is still indexing.
4
Pith papers citing it
17
external citations · OpenAlex
representative citing papers
Latent reasoning faithfulness is a property of training stage and answer format, not of architecture or the final checkpoint alone.
Proposes SCSuff metric for evaluating LLM explanation sufficiency via model-generated alternative inputs, showing explanations are typically insufficient and predictable from hidden states.
Activation verbalization methods for LLMs largely reflect the verbalizer model's parametric knowledge rather than privileged information from the target model's activations.
citing papers explorer
-
Faithfulness Metrics Don't Measure Faithfulness: A Meta-Evaluation with Ground Truth
Introduces BonaFide benchmark of 3,066 ground-truth labeled CoTs showing most faithfulness metrics perform near chance with biases and poor scaling to longer chains.
-
Final Checkpoints Are Not Enough: Analyzing Latent Reasoning Faithfulness Along Training Trajectories
Latent reasoning faithfulness is a property of training stage and answer format, not of architecture or the final checkpoint alone.
-
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.
-
Do Activation Verbalization Methods Convey Privileged Information?
Activation verbalization methods for LLMs largely reflect the verbalizer model's parametric knowledge rather than privileged information from the target model's activations.