Tag
研究人员提出了一种结合深度学习与图像预处理技术的OCR模型，用于自动提取韩文发票中的关键信息，在自建数据集上达到了87%的F1分数，且处理耗时极低。
A user benchmarks the Solar Decide AI model against Jev for handling chat reports in a Roblox game, finding Solar Decide has significantly lower performance and higher false positives, leading to a rollback to Jev.
This study investigates off-target effects of response-style alignment in a Korean 27B language model, finding that post-training for style significantly impacts answer propensity and disclosure rates without targeting safety or capability.
Introduces Ko-WideSearch, a Korean breadth-search benchmark for web agents that evaluates exhaustive set enumeration across 228 tables. Findings show agents have high item recall but struggle with row completion, especially for open-ended cells.
This paper refines word-based grammatical error annotation for L2 Korean by addressing problems in existing resources, including surface target realization and single-reference evaluation, and demonstrates improvements using KoBART-based correction.
NVIDIA's Nemotron-Personas-Korea is a dataset of 6-7 million synthetic personas grounded in official Korean demographic statistics, designed to help build culturally accurate Korean AI agents while complying with Korea's Personal Information Protection Act (PIPA). The tutorial demonstrates how to filter personas and deploy a grounded Korean AI agent using hosted APIs in approximately 20 minutes.
KMMMU is a native Korean benchmark for evaluating multimodal understanding with 3,466 questions across nine disciplines and visual modality categories, addressing the gap of English-centric benchmarks by testing performance on Korean-specific cultural and institutional contexts.