The field of natural language processing is moving towards more efficient and effective methods for evaluating and improving large language models (LLMs). Recent research has focused on developing new metrics and frameworks for assessing LLMs' performance, particularly in tasks such as question answering and mathematical reasoning. One notable direction is the use of natural language inference (NLI) scoring, which has been shown to be a lightweight and effective alternative to more computationally expensive methods. Additionally, there is a growing interest in exploring the limitations and vulnerabilities of LLMs, including their sensitivity to perturbations and their ability to capture human notions of interestingness. The development of new datasets and benchmarks, such as those for Russian information retrieval and mathematical problem generation, is also facilitating progress in these areas. Noteworthy papers include: Revisiting NLI, which demonstrates the effectiveness of NLI-based evaluation for LLMs, and Numerical Sensitivity and Robustness, which highlights the limitations of LLMs in mathematical reasoning tasks. OKBench is also a notable contribution, providing a fully automated framework for generating dynamic knowledge benchmarks. Routesplain is another significant paper, introducing a faithful and intervenable routing approach for software-related tasks.