研究新聞
精選 AIED 期刊與會議論文。每張卡片提供快速概覽，詳情頁則呈現 500 字摘要與實用重點。
免費每日摘要
每日精選研究新聞更新。
單步試用訂閱。目前試用階段不需要付款。
113 篇論文
頁 1 / 19
AIED.HK Editorial
AI Product News Commentary
OpenAI 於 2026 年 9 月 3 日發布 GPT-6 Astra，並引發關於 AGI 是否已經到來的討論。本文將這一說法視為需要歸屬的主張，而非已確立的共識。教育眼前的挑戰，是分清 AI 能產出甚麼，以及學習者能獨立解釋、質疑和遷移甚麼，再以更強的代理能力支援真正的學習。
AIED.HK Editorial
AI Product News Commentary
Anthropic 於 2026 年 9 月 1 日發布 Claude Fable 5.1，強化長程編程與知識工作能力，並降低快取讀取價格。AIED 的機會，是加快從教學構想到可審查原型與研究分析的循環；真正的考驗，是能否把速度轉化為更好的教學與可信證據，同時計入總成本、資料條件和人工審核。
Yearim Kim, Njun Baek, Nojun Kwak
CHI 2026 Workshop on Understanding and Engaging Critical Resistance to AI in Education
Kim, Baek and Kwak present PedaCo, an AI-video workflow that lets educators revise scripts against multimedia-learning principles before synthesis and then checks the rendered video with automated pedagogical metrics. Twenty-three educators rated reviewed videos higher than baselines, while a 14-video comparison improved coherence and temporal alignment. The workshop study did not measure student learning or long-term teacher workload.
OpenAI, Anthropic, Google for Education
AI Product and Learning Report
Product news: ChatGPT now ranks plugin recommendations partly by continued use after installation and adds more time-aware answers, while Claude Code 2.1.239 distinguishes cloud-synced plugins from local installations and makes a data-residency cost premium visible. Gemini for Education supplies the institutional purpose boundary across teaching, learning and work. Together, the updates make tool selection, context, cost and human review part of AI workflow literacy.
Weihao Qu, Ling Zheng, Chris Buzaid, Daniel Crawford
arXiv preprint
Qu and colleagues describe LearnAI, a university framework that paired short course-embedded introductions across 18 courses with optional one-to-one co-creation led by trained undergraduate tutors. Thirty-five clients produced 36 portfolios and more than 20 deployed applications. Interviews and a seven-person readiness comparison are preliminary, so the report supports feasibility and design learning rather than causal claims about skill growth.
Anthropic
AI Product and Learning Report
Product news: Claude Code 2.1.237 introduces a built-in Concise output style and fixes prompt caching for sessions that use an LLM gateway or custom base URL. The release can reduce narration and repeated processing, but brevity and cache efficiency do not establish correctness. Educational teams should preserve task requirements, evidence, tests, and review notes outside the presentation style.