Tag
论文提出 Wieszcz-XIX，一个涵盖 1800–1918 年波兰语的 31 亿词（67.5 亿 token）语料库，并从零训练了 47M 至 349M 参数的时间受限语言模型，展示其在时期词汇上相对现代波兰基座模型的交叉优势，同时发布语料、代码与权重。
This paper introduces a Polish medical visual question answering benchmark built from Polish Board Certification Examination questions, evaluating vision-language models and showing that they underutilize visual evidence compared to question text.
This paper introduces a new benchmark for evaluating LLMs on Polish high school history exit exams (Matura), showing that models outperform humans on aggregate scores but exhibit distinct failure modes in source interpretation and temporal reasoning.
Introduces PoVisLE, a Polish vision-language evaluation benchmark with 1,117 images and 2,366 manually annotated VQA pairs, designed to assess culturally grounded multimodal understanding beyond surface-level recognition.
A detailed Twitter thread outlining 10 rules for achieving polished UI with Claude, covering easing curves, design system variables, physics-based animations, and other refinements.
This paper investigates intrinsic data metrics to predict the utility of reasoning supervision before costly fine-tuning, finding that smaller models benefit from alignment-focused metrics while larger models gain from verbose traces, thus establishing a scale-aware framework for validating reasoning datasets.