Selection, Reflection and Self-Refinement: Revisit Reasoning Tasks via a Causal Lens
arXiv:2510.08222v2 Announce Type: replace
Abstract: Due to their inherent complexity, reasoning tasks have long been regarded as rigorous benchmarks for assessing the capabilities of machine learning models, especially large language models (LLMs). Al…