Dynamically adjusting beta via LLM-as-judge downweights biased comparisons to learn more rational reward models from flawed human preferences.
2410.15413 , primaryclass =
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
fields
cs.AI 2verdicts
UNVERDICTED 2representative citing papers
LLMs exhibit quality-dependent order biases and name biases in pairwise comparisons that can cause selection of inferior options, demonstrated across resume and color tasks with a new classification of preferences as robust, fragile, or indifferent.
citing papers explorer
-
Mitigating Cognitive Bias in RLHF by Altering Rationality
Dynamically adjusting beta via LLM-as-judge downweights biased comparisons to learn more rational reward models from flawed human preferences.
-
Fragile Preferences: A Deep Dive Into Order Effects in Large Language Models
LLMs exhibit quality-dependent order biases and name biases in pairwise comparisons that can cause selection of inferior options, demonstrated across resume and color tasks with a new classification of preferences as robust, fragile, or indifferent.