Breadcrumb
- Home
- AI Hub For Education
- Research Study Repository
- Evidence Explorer
- AI Tools
- Gemini / Gemma
The Repository
Search and Filter
Gemini / Gemma
Studies tagged with AI from their full text or abstract; see each study for details.
Showing 211 - 240 of 326 studies
Visiomath: Benchmarking Figure-Based Mathematical Reasoning In Lmms
Can Li, Ying Liu, Ting Zhang, Mei Wang, Hua Huang. (10/2025). arXiv.
Multimodal models showed consistent accuracy declines as visual similarity among diagrammatic choices increased, with top models like Gemini 2.5 Pro reaching 80.9% overall accuracy compared to 91.3% for humans.
Beyond The Benefits: A Systematic Review Of The Harms And Consequences Of Generative Ai In Computing Education
Seth Bernstein, Ashfin Rahman, Nadia Sharifi, Ariunjargal Terbish, Stephen MacNeil. (10/2025). arXiv.
Cognitive harms were the most prevalent category across the reviewed papers (32.4%) and had the highest proportion supported by empirical evidence (28.2%).
Teachlm: Post-Training Llms For Education Using Authentic Learning Data
Janos Perczel, Jin Chow, Dorottya Demszky. (10/2025). arXiv.
Fine-tuning models on authentic tutoring data improved pedagogical performance, doubling student talk time and reducing tutor verbosity.
Seephys: Does Seeing Help Thinking? Benchmarking Vision-Based Physics Reasoning
Kun Xiang, Heng Li, Terry Jingchen Zhang, Yinya Huang, Zirong Liu, Peixin Qu, Jixi He, Jiaqi Chen, Yu-Jie Yuan, Jianhua Han, Hang Xu, Hanhui Li, Mrinmaya Sachan, Xiaodan Liang. (10/2025). arXiv.
Even top-performing models struggle on the benchmark, with Gemini-2.5-Pro achieving 54.9% accuracy and OpenAI o4-mini reaching 51.9%, well below human experts at 86.5%.
Exploring Student Choice And The Use Of Multimodal Generative Ai In Programming Learning
Xinying Hou, Ruiwei Xiao, Runlong Ye, Michael Liut, John Stamper. (10/2025). arXiv.
Students most frequently chose screen-sharing as their context input modality (18 cases), pairing it with speech input in 83% of those cases and audio output in 84%.
Geolog-Ia: Sistema Conversacional Sobre Tesis AcadéMicas
Micaela Fuel Pozo, Andrea Guatumillo Saltos, Yeseña Tipan Llumiquinga, Kelly Lascano Aguirre, Marilyn Castillo Jara, Christian Mejía-Escobar. (10/2025). arXiv.
Geolog-IA achieved an average BLEU score of 0.87 when comparing its generated answers to actual database query results.
Editlens: Quantifying The Extent Of Ai Editing In Text
Katherine Thai, Bradley Emi, Elyas Masrour, Mohit Iyyer. (10/2025). arXiv.
EDITLENS achieved 94.0% accuracy in binary classification and 90.2% accuracy in ternary classification, outperforming existing detection baselines.
Lightweight Prompt Engineering For Cognitive Alignment In Educational Ai: A Oneclickquiz Case Study
Antoun Yaacoub, Jerome Da-Rugna, Zainab Assaghir. (10/2025). arXiv.
Detailed prompts with explicit level descriptions and action verbs achieved a 96% match rate to intended Bloom's levels, significantly outperforming simpler and persona-based prompts.
Enhancing Personalized Multi-Turn Dialogue With Curiosity Reward
Yanming Wan, Jiaxing Wu, Marwa Abdulhai, Lior Shani, Natasha Jaques. (10/2025). arXiv.
Incorporating curiosity rewards based on differential log accuracy led human evaluators to prefer the model's personalization 75.75% of the time over standard multi-turn RLHF in education dialogues.
Code2Video: A Code-Centric Paradigm For Educational Video Generation
Yanzhe Chen, Kevin Qinghong Lin, Mike Zheng Shou. (10/2025). arXiv.
Code2Video achieved a 40% improvement in teaching effectiveness over direct code generation and substantially outperformed pixel-based video models.
A Meta-Analysis Of Llm Effects On Students Across Qualification, Socialisation, And Subjectification
Jiayu Huang, Ruoxin Ritter Wang, Jen-Hao Liu, Boming Xia, Yue Huang, Ruoxi Sun, Jason (Minhui) Xue, Jinan Zou. (09/2025). arXiv.
Large language models showed significant positive overall effects on academic qualification (g = 0.751), socialisation (g = 0.745), and learner subjectification (g = 0.654).
Towards An Ai-Augmented Textbook
LearnLM Team. (09/2025). arXiv.
Students using the AI-augmented textbook scored significantly higher on both immediate (77% vs. 68%) and three-day retention assessments (78% vs. 67%) compared to digital reader users.
LLMS4ALL: A Review On Large Language Models For Research And Applications In Academic Disciplines
Yanfang (Fanny) Ye*, Zheyuan Zhang, Tianyi Ma, Zehong Wang, Yiyang Li, Shifu Hou, Weixiang Sun, Kaiwen Shi, Yijun Ma, Wei Song, Ahmed Abbasi, Ying Cheng, Jane Cleland-Huang, Steven Corcelli, Robert Goulding, Ming Hu, Ting Hua, John Lalor, Fang Liu, Tengfei Luo, Ed Maginn, Nuno Moniz, Jason Rohr, Brett Savoie, Daniel Slate, Tom Stapleford, Matthew Webber, Olaf Wiest, Johnny Zhang, Nitesh V Chawla. (09/2025). arXiv.
No single model dominates across all domains, as general conversational systems excel on dialogue while specialized or reasoning models perform best on mathematics, code, and structured logic.
Developing Strategies to Increase Capacity in AI Education
Noah Q. Cowit, Sri Yash Tadimalla, Stephanie T. Jones, Mary Lou Maher, Tracy Camp, Enrico Pontelli. (09/2025). arXiv.
A shortage of faculty with AI expertise and a lack of compensated time for training present major obstacles to expanding courses.
A Meta-Analysis Of LLM Effects On Students Across Qualification, Socialization, And Subjectification
Jiayu Huang, Ruoxin Ritter Wang, Jen-Hao Liu, Boming Xia, Yue Huang, Ruoxi Sun, Jason (Minhui) Xue, Jinan Zou. (09/2025). arXiv.
Large language models showed significant positive overall effects on qualification (g = 0.751), socialization (g = 0.745), and learner subjectification (g = 0.654), though subjectification was the least stable.
A Mega-Study Of Digital Twins Reveals Strengths, Weaknesses And Opportunities For Further Improvement
Tiany Peng, George Gui, Daniel J. Merlau, Grace Jiarui Fan, Malek Ben Sliman, Melanie Brucks, Eric J. Johnson, Vicki Morwitz, Abdullah Althenayyan, Silvia Bellezza, Dante Donati, Hortense Fong, Elizabeth Friedman, Ariana Guevara, Mohamed Hussein, Kinshuk Jerath, Bruce Kogut, Kristen Lane, Hannah Li, Patryk Perkowski, Oded Netzer, Olivier Toubia. (09/2025). arXiv.
Digital twins' answers were only modestly more accurate than generic base language models and correlated weakly with actual human choices (average correlation of 0.20).
ClassMind: Scaling Classroom Observation and Instructional Feedback with Multimodal AI
Ao Qu, Yuxi Wen, Jiayi Zhang, Yunge Wen, Yibo Zhao, Alok Prakash, Andres Salazar-Gomez, Paul Pu Liang, Jinhua Zhao. (09/2025). arXiv.
The AVA-Align video agent achieved 100% factuality and higher significance and feasibility scores than Gemini-2.5-Pro and Deep Video Discovery when evaluated on ten classroom videos.
AI Academy: Building Generative AI Literacy in Higher Ed Instructors
Si Chen, Xiuxiu Tang, Alison Cheng, Nitesh V Chawla, G. Alex Ambrose, Ronald Metoyer. (09/2025). arXiv.
Instructors showed significant self-assessed gains in six of nine generative AI literacy dimensions, including assessing outputs, prompt engineering, and tool use.
Prompting the Professoriate: A Qualitative Study of Instructor Perspectives on LLMs in Data Science Education*
Ana Elisa Lopez-Miranda, Tiffany Timbers, Rohan Alexander. (09/2025). arXiv.
Most instructors held pragmatic views toward language models, but 31 percent had not used them to teach and had no plans to do so.
Bridging Cultural Distance Between Models Default and Local Classroom Demands: How Global Teachers Adopt GenAI to Support Everyday Teaching Practices
Ruiwei Xiao, Qing Xiao, Xinying Hou, Hanqi Li, Phenyo Phemelo Moletsane, Hong Shen, John Stamper. (09/2025). arXiv.
Cultural distance operates across three levels: low distance requiring minor edits, mid distance needing sustained prompting and manual revision, and high distance presenting unbridgeable structural barriers.
Emerging Patterns of GenAI Use in K-12 Science and Mathematics Education
Lief Esbenshade, Shawon Sarkar, Drew Nucci, Ann Edwards, Sarah Nielsen, Joshua M. Rosenberg, Alex Liu, Zewei Tian, Min Sun, Zachary Zhang, Thomas Han, Yulia Lapicus, Kevin He. (09/2025). arXiv.
Over 50% of surveyed math and science teachers reported using generative AI tools, with 22% using them at least weekly.
Prompting Strategies for Language Model-Based Item Generation in K-12 Education: Bridging the Gap Between Small and Large Language Models
Mohammad Amini, Babak Ahmadi, Xiaomeng Xiong, Yilin Zhang, Christopher Qiao. (08/2025). arXiv.
Under human and simulated expert scoring, the fine-tuned Gemma model achieved a higher overall average score (3.90 out of 5) than off-the-shelf GPT-3.5 (3.34).
MathBuddy: A Multimodal System for Affective Math Tutoring
Debanjana Kar, Leopold Böss, Dacia Braca, Sebastian Maximilian Dennerlein, Nina Christine Hubig, Philipp Wintersberger, Yufang Hou. (08/2025). arXiv.
Integrating student emotion features into the LearnLM math tutor increased its pedagogical win rate against ground truth responses by 23 percentage points.
Reliable generation of isomorphic physics problems using ChatGPT with prompt-chaining and tool use
Zhongzhou Chen. (08/2025). arXiv.
Prompt chaining combined with the Python code interpreter produced consistent physics problems with verified solutions and usable diagrams, whereas simpler prompts failed or produced calculation hallucinations.
Enabling Multi-Agent Systems as Learning Designers: Applying Learning Sciences to AI Instructional Design
Jiayi Wang, Ruiwei Xiao, Xinying Hou, John Stamper. (08/2025). arXiv.
Teachers strongly preferred activities from the collaborative multi-agent discussion system, rating them as significantly more creative, contextually relevant, and classroom-ready.
Teaching Introduction to Programming in the Times of AI: A Case Study of a Course Redesign
Nikolaos Avouris, Kyriakos Sgarbas, George Caridakis, Christos Sintoris. (08/2025). arXiv.
ChatGPT o3-mini-high scored 38.4 out of 40 on course assignments without follow-up prompts, placing it in the top 10% of enrolled students.
Too Easily Fooled? Prompt Injection Breaks LLMs on Frustratingly Simple Multiple-Choice Questions
Xuyang Guo, Zekai Huang, Zhao Song, Jiahao Zhang. (08/2025). arXiv.
Visible black-text injection prompts successfully misled all tested models into choosing incorrect answers, even on trivial arithmetic questions.
From Misunderstandings To Learning Opportunities: Leveraging Generative AI In Discussion Forums To Support Student Learning
Stanislav Pozdniakov, Jonathan Brazil, Oleksandra Poquet, Stephan Krusche, Santiago Berrezueta-Guzman, Shazia Sadiq, Hassan Khosravi. (08/2025). arXiv.
All five instructors supported the M2M approach, viewing AI-identified student misunderstandings and generated learning opportunities as insightful, actionable, and valuable for weekly teaching.
CODAE: Adapting Large Language Models for Education via Chain-of-Thought Data Augmentation
Shuzhou Yuan, William LaCroix, Hardik Ghoshal, Ercong Nie, Michael Färber. (08/2025). arXiv.
Models fine-tuned on the combined augmented data delivered higher pedagogical helpfulness and scaffolding scores, reaching up to 4.70 out of 5 for helpfulness in Gemma 2.
A Fuzzy Logic Prompting Framework for Large Language Models in Adaptive and Uncertain Tasks
Vanessa Figueiredo. (08/2025). arXiv.
Scaffolded prompts led to significantly higher scaffolding quality scores across models compared to standard prompting baselines.
