The field of large language models (LLMs) is moving towards improving reliability and trustworthiness. Recent research has focused on developing methods to estimate LLM consistency, uncertainty, and confidence. Studies have shown that existing approaches to measuring consistency and uncertainty often do not align with human perceptions, highlighting the need for more robust evaluation methods. Additionally, the use of rationales and premature layer interpolation have been explored as ways to enhance factuality and reduce hallucinations in LLMs. Noteworthy papers in this area include: Estimating LLM Consistency, which proposes a logit-based ensemble method for estimating LLM consistency. Read Your Own Mind, which shows that reliable uncertainty estimation requires explicit exploration of the generative space. Reinforcement Learning for Better Verbalized Confidence, which introduces a novel verbalized confidence estimation method for long-form generation. Should I Share this Translation, which evaluates quality feedback for user reliance on machine translation. Expanding before Inferring, which proposes a novel intervention to enhance factuality through premature layer interpolation. Verbalized Confidence Triggers Self-Verification, which finds that supervised fine-tuning with scalar confidence labels alone suffices to elicit self-verification behavior. High Accuracy, Less Talk (HALT), which proposes post-training an LLM to generate content only when confident in its correctness. MetaFaith, which introduces a novel prompt-based calibration approach inspired by human metacognition to improve faithful calibration of LLMs.