A recent study revisits Contrast-Consistent Search (CCS), an unsupervised probing method designed for large language models, with the goal of improving understanding and applicability of this technique. The research specifically focuses on clarifying the mechanisms underlying CCS and enhancing its performance by optimizing relative contrast. This optimization aims to better capture how models represent binary features, such as sentence truth. By refining CCS, the study contributes to deeper insights into the internal representations of language models and their interpretability. The work also highlights the potential for broader application of CCS in analyzing model behavior. Overall, this research advances the methodological toolkit available for probing large language models, promising improved analysis of their learned features.