Open Ko-LLM Leaderboard: Evaluating Large Language Models in Korean with Ko-H5 Benchmark

This paper introduces the Open Ko-LLM Leaderboard and the Ko-H5 Benchmark as vital tools for evaluating Large Language Models (LLMs) in Korean. Incorporating private test sets while mirroring the English Open LLM Leaderboard, we establish a robust evaluation framework that has been well integrated in the Korean LLM community. We perform data leakage analysis that shows the benefit of private test sets along with a correlation study within the Ko-H5 benchmark and temporal analyses of the Ko-H5 score. Moreover, we present empirical support for the need to expand beyond set benchmarks. We hope the Open Ko-LLM Leaderboard sets precedent for expanding LLM evaluation to foster more linguistic diversity.

Data-Driven Web-Based Patching Management Tool Using Multi-Sensor Pavement Structure Measurements

Automating pavement maintenance suggestions is challenging,especially for actionable recommendations such as patching location,depth and priority.It is common practice among State agencies to manually inspect road segments of interest and decide maintenance requirements based on the pavement condition index (PCI).However,standalone PCI only evaluates the pavement surface condition and coupled with the variability in human perception of pavement distress,limits the accuracy and quality of current pavement maintenance practices.Here,a need for multi-sensor data integrated with standardized pavement distress condition ratings is required.This study explores the possibility of estimating the appropriate pavement patching strategy (i.e.,patching location,depth,and quantity) by integrating pavement structural and surface condition assessment with pavement specific ratings of distress.Especially,it combines pavement structural condition assessment parameter;falling weight deflectometer deflections along with surface condition assessment parameters;international roughness index,and cracking density for a better representation of overall pavement distress condition.Then,a pavement specific threshold-based patching suggestion algorithm is implemented to evaluate the pavement overall distress condition into a priority-based patching suggestion.The novelty in the use of pavement specific thresholds is placed on its data-driven ability to determine threshold values from current road condition measurements using a reliability concept validated by the theoretical pavement condition rating,pavement structural number.A web-based patching manager tool (PMT) was implemented to automate the patching suggestion procedure and visualize the results.Validated with road surface images obtained from three-dimensional laser sensors,PMT could successfully capture localized distresses in existing pavements.