본 연구는 Deepseek와 ChatGPT가 중국 대학 한국어 학습자들의 쓰 기 텍스트에 대해 제공하는 피드백 양상을 살피고, AI 피드백의 활용 방 안을 모색하는 데 목적이 있다. 이를 위해 한국어 학습자 30명을 대상으 로 쓰기 텍스트를 수집하고, 두 도구가 텍스트에 대해 제공한 피드백을 분석하였다. 분석 결과, 두 도구 모두 평가 기준에 기반하여 피드백을 제 공하고, 학습자가 수용하기 쉽도록 문제점과 개선 방향을 명확하게 제시 하였다. 대부분의 피드백은 정확도가 높은 편이었으나, 평가 범주와 오류 유형이 부합하지 않는 경우도 일부 나타나고 있었고, 특히 Deepseek의 경우 쓰기 주제를 잘못 파악하여 내용적 측면에서 부적절한 피드백을 제 공한 사례가 많았다. 두 도구 모두 쓰기 텍스트의 내용에 따라 피드백을 제공하여 우선순위는 두드러지지 않았고, Deepseek가 ChatGPT에 비해 지지적인 어조를 더 많이 사용하는 것으로 나타났다. 본 연구는 생성형 AI를 활용한 한국어 교육 및 연구를 위해 기초적인 자료로 사용할 수 있 을 것이다.

본 연구는 Chat GPT 기반으로 개발한 실험자용 생성형 AI 코칭 시스템 ‘Roh-KAI(로카이)’가 인간 사용자 간 상호작용과정에서 발생하는 사회적 존재 감, 신뢰, 자기효능감의 심리적 변화를 탐색하고자 하였다. 로카이는 국제코칭연 맹(ICF)과 한국코치협회(KCA)의 윤리규정 및 핵심 역량을 반영하여 8단계 47개 질문 구조로 설계되었다. 인공지능 사용경험이 있는 성인 15명을 대상으로 주제 분석 기법을 적용한 질적 연구를 진행하였다. 분석 결과 4개의 상위 주제와 9개의 하위 주제가 도출되었다. 사회적 존재감 에서는 60%가 대화의 자연스러움을, 80%가 공감적 반응을 경험했고, 인공지능 신뢰에서는 53.3%가 공감적 경청과 이해를, 40%가 비판단적 태도를 인식했다. 자기효능감에서는 66.7%가 자기 인식 확장을, 46.7%가 실행 의지 강화를 나타 냈다. 반면 53.3%가 구조화된 프로세스의 제약을 지적했고, 33.3%가 사용자 다 양성 대응 부족을, 26.7%가 윤리적 우려를 지적하였다. 연구 결과 AI 코칭 시스 템이 공감적 상호작용과 자기성찰 촉진에는 효과적이나, 시스템의 유연성과 개 별화된 접근의 중요성을 확인하였다. 본 연구는 AI 코칭 시스템의 설계와 개선 방향에 실증적 근거를 제공하는 데 의의가 있다.

This study proposes a real-time content design pipeline optimized for Unreal Engine, integrating generative AI-based image creation with AI-assisted 3D modeling tools. The pipeline aims to streamline the production of high-quality assets for real-time applications, including games and simulations. Two types of subjects were selected: a bust combining organic character features, and a stone slab characterized by planar and symmetrical structure. Multi-angle image data were first synthesized using advanced generative AI models to simulate diverse viewpoints. These were then processed using AI-enhanced photogrammetry and modeling tools to reconstruct detailed 3D meshes and extract base textures. Post-processing steps, including mesh decimation, UV unwrapping, and texture baking, were performed to ensure compatibility with Physically Based Rendering (PBR) workflows used in Unreal Engine. The final assets were successfully imported into Unreal Engine, demonstrating visual fidelity and performance suitability in a real-time environment. The study confirms the pipeline’s potential for accelerating asset development and suggests promising future directions in AI-driven digital content creation.

This study structurally analyzes the algorithmic filtering process by which generative AI images are either selected or discarded before reaching users, and models this process through a visual similarity–based simulation. Images generated by Stable Diffusion are placed on a two-dimensional grid, and a modified version of Conway’s Game of Life algorithm is applied to update the state of each cell. The survival of each cell is determined based on a hybrid visual similarity metric combining CLIP and LPIPS. To prevent the rigidity of the simulation and sustain emergent dynamics, random image injections are periodically introduced. The simulation results reveal that visually similar images repeatedly form clusters, and a visual order gradually converges toward a structurally stabilized state. This suggests that specific visual orders can emerge solely from algorithmic selection criteria, independent of human interpretation. By shifting focus from semantic or symbolic analysis to the experimental conditions for the existence and persistence of images, this study proposes a new analytical perspective for understanding digital image environments.

오늘날 생성형 인공지능(Generative AI)의 비약적 발전은 교육 패러다 임에 근본적인 변화를 초래하고 있으며, 특히 ChatGPT(Generative Pre-trained Transformer)와 같은 대화형 AI는 단순한 정보 전달을 넘 어, 상호작용 기반의 학습 환경을 가능하게 한다는 점에서 언어교육 분 야에서 주목받고 있다. 본 연구는 생성형 인공지능의 대표 사례인 ChatGPT를 일본어 문법 교육에 활용하였을 때의 교육적 기능과 잠재적 한계를 분석하는 데 목적이 있다. 연구는 일본어를 학습한 지 3개월 정 도 지난 초급 학습자를 대상으로, ChatGPT의 문법적 설명 능력, 예문 생성력, 피드백의 적절성에 초점을 맞추어 실험을 설계하였다. 구체적으 로 사전·사후 문법 이해도 테스트, 만족도 설문조사, 자유 서술 및 인터 뷰를 통해 데이터를 수집하고 정량적·정성적으로 분석하였다. 그 결과 예 문 제공, 반복 질문 가능성, 심리적 부담 감소 등의 장점을 긍정적으로 평가하였다. 반면, 설명의 일관성 부족, 문맥 판단력의 한계, 정교한 문 법 개념 설명의 부재 등은 주요 한계로 지적되었다. 본 연구는 ChatGPT 가 일본어 문법 교육에서 자기주도 학습을 촉진하는 보조 도구로서 효과 적일 수 있음을 시사하며, 동시에 문법 교육의 질을 담보하기 위해서는 교 사의 맥락적 설명과 피드백이 반드시 병행되어야 함을 강조한다. 본 연구 가 향후 생성형AI와 교사의 역할을 상호 보완적으로 융합한 혼합형 (hybird) 교육 모델의 기초 자료가 되길 기대한다.

This study examined how 16 Chinese transformational structures are generated using generative AI from the perspective of learners whose native language is Korean. To summarize: (1) In weak AI models, using the zero-shot input method, Baidu generated 13 transformed Chinese Sentences, and Papago generated 11 transformed Chinese Sentences. (2) In strong AI models, using the prompt input method, WRTN generated 12 transformed Chinese Sentences, and Yuanbao generated 11 transformed Chinese Sentences. The possible reason why weak AI showed better results than strong AI may be because the analysis target was simple sentences. Baidu and Papago AI are programs specialized in translation. Therefore, under the same conditions as the experiment, it can posited that weak AI is more specialized than strong AI. Thus, it may be sufficient to utilize weak AI in current Chinese writing education. Nevertheless, for this research be applicable to Chinese writing education, the following additional analyses are necessary: (1) This study targeted ‘simple sentences.’ If applied to ‘complex sentence’ writing education, an analysis of whether weak AI remains useful is necessary. (2) An analysis of how to conduct education using Artificial Intelligence is required.

This study evaluates how three AI models—ChatGPT, DeepSeek, and Clova X—detect and respond to gender-biased expressions in Korean. Clova X exhibited the highest accuracy in identifying discriminatory language, followed by ChatGPT, while DeepSeek performed the poorest. While terms like “kimchi girl” and “doenjang girl” were correctly recognized, phrases such as “female doctor” and “maiden work” were often misinterpreted. ChatGPT and DeepSeek occasionally provided inaccurate definitions, raising concerns about misinformation. Interestingly, DeepSeek performed best when interpreting sexist proverbs, although the overall differences across models were minor. All three models generally succeeded in recognizing biased expressions in conversational contexts, but DeepSeek struggled with non-standard sentence formats, leading to delays or missing responses. These results reveal current limitations in generative AI’s ability to process culturally specific and nuanced language. This study emphasized the need to incorporate more diverse Korean language data and up-to-date linguistic research in AI training. As generative AI tools become more integrated into everyday communication, improving their ability to detect and respond to gender biases is crucial for fostering fair and responsible language technologies.

The present study investigates the use of generative artificial intelligence (AI) tools by pre-service teachers (PSTs) in lesson planning for a middle-school English as a foreign language (EFL) class, aiming to address gaps and inform teacher training. The case study examined PSTs in a South Korean university course who were tasked with creating lesson plans using generative AI to aid in lesson plan development for a middle school lesson that incorporated generative AI. Data were analyzed thematically, and results revealed that generative AI was used in topic selection, material creation, lesson organization, and language checking. While generative AI facilitated efficiency and creativity, challenges emerged, including the quality of outputs and limited incorporation of effective pedagogical strategies. These findings indicate a need for targeted training in prompt engineering, ethical considerations, pedagogy, and collaborative practices to enhance PSTs’ generative AI competencies. This study contributes to teacher education programs by providing insights into the practical integration of generative AI in pedagogical practices.

Advances in digital tools and building structure technologies have enabled more flexible architectural design, with AI-based performance design gaining considerable attention as a new design methodology. Stadium design must consider the two primary elements of sports events: athletes and spectators. Given that the facade of a stadium directly impacts solar energy efficiency, it is essential to incorporate environmental performance considerations from the initial design phase. This study employs an AI-based Generative Design process to generate a facade form that efficiently manages solar radiation and daylight, satisfying two conflicting performance objectives: max- imizing sunlight for turf growth in the pitch zone and minimizing direct sunlight exposure in the stadium seating zone. The optimal solution derived ranks 331st for pitch zone sunlight and 408th for stadium seating sunlight out of a dataset of 1,000 models. While this solution does not represent the absolute best for either individual objective, it is evaluated as the most balanced alternative, achieving the goal of maximizing sunlight in the pitch zone and minimizing it in the seating zone

This study proposes and verifies a next-generation 3D modeling pipeline integrating Generative AI and Photogrammetry techniques. Two types of objects were selected for experimentation: a bust combining characteristics of both objects and characters, and a stone slab exhibiting planar and symmetrical features. Firstly, videos depicting these objects from various viewpoints were generated using generative AI models. Secondly, these videos were processed through Photogrammetry software to produce high-quality 3D meshes and diffuse textures. Lastly, post-processing including high and Low-polygon mesh separation, UV unwrapping, and texture baking was performed, successfully producing Physically Based Rendering (PBR)-compatible game assets. The proposed pipeline demonstrates significant potential for practical applications across various industries, and further advances in generative AI and future research are expected to broaden this novel approach to digital asset creation.