Are Implicit Reasoning Models Really Hard to Explain? A Deep Study on the Interpretability of LRMs

Introduction / Main Post: Are Implicit Reasoning Models Really Hard to Explain? A Deep Study on the Interpretability of LRMs

Through empirical research, this paper finds that the reasoning tokens of implicit reasoning models are often not necessary, and in most cases, interpretable natural language reasoning trajectories can be decoded. This indicates that current LRMs actually encode interpretable processes, and interpretability itself can serve as a signal for predicting correctness.