Are LLMs Positionally Consistent Ordinal Classifiers? A Systematic Evaluation

Excerpt

arXiv:2608.08869v2 Announce Type: replace Abstract: Large language models are increasingly used for ordinal classification, yet semantically equivalent changes to prompt organization can alter their predictions. We conduct systematic experiments to characterize positional bias from label order, demonstration order, and demonstration placement. First, we apply the three probes to ten frontier LLMs on a common ordinal-classification task; every model is sensitive to all three positional sources, s