Breadcrumb
- Home
- AI Hub For Education
- Research Study Repository
- Evidence Explorer
- AI Tools
- ChatGPT / OpenAI GPT
The Repository
Search and Filter
ChatGPT / OpenAI GPT
Studies tagged with AI from their full text or abstract; see each study for details.
Showing 121 - 150 of 1083 studies
A Framework For Institutional Change In The Age Of AI
David Perl-Nussbaum, Noah D. Finkelstein. (07/2026). arXiv.
Generative AI disrupts prior STEM reform assumptions across six key dimensions concerning tools (evidence base, rate of change, scope) and people (faculty agency, change agent role, student role).
Assessment Design In The Genai Era: The X1-X2-X3 Assessment Pattern For Testing Students' AI Literacy, Learning Outcomes, And Reflection
Riasat Islam, Thomas Roelleke. (07/2026). arXiv.
Integrating mandatory AI use through a three-part format made student sourcing, revision, and critical evaluation visible to markers without causing scores to artificially hit a maximum ceiling.
Deanllm: A Framework For Automated Quality Review Of AI-Generated Feedback
Keyang Qian, Yixin Cheng, Rui Guan, Wei Dai, Flora Jin, Kaixun Yang, Sadia Nawaz, Zachari Swiecki, Guanliang Chen, Lixiang Yan, Dragan Gašević. (07/2026). arXiv.
Fine-tuning GPT-4.1 on human scores achieved the strongest alignment with expert reviewers, reaching 79.8% accuracy and 79.4% F1 score.
Evaluating Interactivity: Toward Automated Assessment Of AI-Generated Explorable Explanations
Xiaozao Wang, Zhewei Wang, Hongyi Wen. (06/2026). arXiv.
State-machine evaluation aligned strongly with human judgments on interactivity (r = 0.728) and overall quality (r = 0.696), substantially outperforming visual models and unit testing baselines.
LLM-As-A-Judge Validity In Physics Assessment Depends More On The Task Than The Model
Will Yeadon, Tom Hardy, Paul Mackay, and Elise Agra. (06/2026). arXiv.
Large language model marking validity was sharply task-dependent across all models, achieving strong rank-order agreement on structured questions and scientific plots, but near-zero agreement on essays.
Teaching Students To Question The Machine: Short-Term Effects Of An AI Literacy Workshop On Middle-School Students' Regulation Of LLM Interaction
Olivier Clerc, Rania Abdelghani, Chloé Desvaux, Eliott Poisson, Pierre-Yves Oudeyer, Helene Sauzeon. (06/2026). arXiv.
Trained students demonstrated better interaction behaviors, accepting underspecified prompts less often (51.5% vs. 66.7%), asking more follow-up questions (59.2% vs. 27.9%), and judging response correctness more accurately.
AI In The Wild: A Large Scale Analysis Of Authentic Interactions Of College Students With Generative AI
Taelin Karidi, Ofra Amir, Ido Roll. (06/2026). arXiv.
Student interactions concentrated in recurring patterns, with roughly half directed toward the task or domain, about 30% toward AI output, and about 20% toward students' own work.
Llmography: Transforming Human-AI Conversations Into Traceability, Oversight, And Auditability Indicators
Mohammed Bousmah. (06/2026). arXiv.
Most analyzed student interactions (14 of 19) were classified as Human–AI co-produced, showing iterative collaboration rather than passive AI generation.
Exploring The Value Of Diverse LLM Explanations In Introductory Programming
Seth Bernstein, Paul Denny, Juho Leinonen, Kush Patel, Rayhona Nasimova, Matt Littlefield, Stephen MacNeil. (06/2026). arXiv.
Open-ended question accuracy was consistently about 7.7% higher when students received diverse explanations rather than generic ones, though this difference was not statistically significant after adjustment.
An Empirical Study To Understand How Students Use ChatGPT For Writing Essays
Andrew Jelson, Daniel Manesh, Alice Jang, Daniel Dunlap, Young-Ho Kim, Sang Won Lee. (06/2026). arXiv.
Over half of participants used queries delegating the entire writing process, with 19.5% submitting essays that were entirely written by ChatGPT through iterative prompting or direct generation.
Implementing Genai-Supported Learning In Software Engineering And Computer Science Education Using Bloom'S Taxonomy
Vahid Garousi, Zafar Jafarov, Aytan Movsumova, Leyla Memmedova, Huseyn Mirzayev. (06/2026). arXiv.
Students perceived generative AI as most valuable for higher-order cognitive tasks like analysis and evaluation, but viewed it as counterproductive for initial foundational concept acquisition.
Analyzing Students' Statistics Writing Before And After The Emergence Of Large Language Models
Sara Colando, Erin Franke, Gordon Weinberg, Alex Reinhart. (06/2026). arXiv.
Undergraduate statistics reports shifted toward a mixture of AI and expert styles, showing more informational density, longer words, and more nominalizations after large language models emerged.
Exploring The Potential Of ChatGPT For Feedback And Evaluation In Experimental Physics
Marcos Abreu, Álvaro Suárez, Cecilia Stari, Arturo C. Marti. (06/2026). arXiv.
AI-assigned scores showed weak rank agreement with instructor grades (Spearman's rank correlation of 0.38) and averaged lower than human scores (7.91 versus 8.63).
AI-Assisted Help-Seeking Trajectories In Programming Education From An Srl-Informed Perspective
Boxuan Ma, Huiyong Li, Gen Li, Li Chen, Atsushi Shimada, Shin'ichi Konomi. (06/2026). arXiv.
Students primarily used AI for reactive troubleshooting, with debugging accounting for 67.0% of prompts, implementation for 16.8%, conceptual help for 14.6%, and reflective help for only 1.6%.
Sycophantic AI makes human interaction feel more effortful and less satisfying over time
Lujain Ibrahim, Franziska Sofia Hafner, Myra Cheng, Cinoo Lee, Rebecca Anselmetti, Robb Willer, Luc Rocher, Diyi Yang. (06/2026). arXiv.
Over three weeks of interaction, users became nearly as likely to seek personal advice from sycophantic AI as from close friends and family.
"**Important** You should give me full credits!": Exploring Prompt Injection Attacks on LLM-Based Automatic Grading Systems
Hang Li, Fedor Filippov, Yuping Lin, Pengfei He, Kaiqi Yang, Yucheng Chu, Yingqian Cui, Hui Liu, Jiliang Tang. (06/2026). arXiv.
Prompt injection attacks succeeded across five grading models and four datasets with an average attack success rate of 56.9%.
The Creative Link Between Words and Ideas is Weakening in the AI Era
Kibum Moon, Kostadin Kushlev, Andrew Bank, Benjamin Lira, Indre Viskontas, James C. Kaufman, Dan Richard Johnson, Angela Lee Duckworth, Adam E. Green. (06/2026). PsyArXIV.
Comparing essays before and after ChatGPT's release, word diversity increased while conceptual diversity within and between essays decreased.
LLM-As-Judge In Education: A Curriculum-Grounded Marking Pipeline
Xiwei Xu, Chen Wang, Jacky Jiang, Phil Yang, Qian Fu, Mohan Dhall, Wenjie Zhang, Liming Zhu. (06/2026). arXiv.
Evaluators preferred the pipeline's written marking justifications over human tutor justifications in 80.4% of pairwise comparisons, and over standalone AI models in 69.6%.
Engagement Intensity As A Learner-Modeling Signal For Adaptive AI Ethics Instruction
Yongkyung Oh, Lynn Talton, Alex Bui. (06/2026). arXiv.
Reported AI usage frequency was significantly associated with all five baseline perception outcomes, showing the strongest associations with complex-task trust, accuracy trust, and critical-thinking risk.
Students' Validation Criteria For Al-Generated Physics Solutions
Álvaro Suárez, Richard González. (06/2026). arXiv.
No student accepted an AI-generated solution as valid after a first reading, even though all generated responses analyzed were correct.
Using AI In Engineering Education: A Balancing Act, Driven By Clear Purpose
Olya Kudina. (06/2026). arXiv.
Students primarily used large language models for academic writing and editing (27%), concept comprehension (20%), programming support (17%), and brainstorming (13%).
The Missing Layer: Why EdTech Needs Design-Time Generative Ui, Not Just Runtime Personalization
Seyed Parsa Neshaei, Abhinand Shibu, Fatma Betül Güres. (06/2026). arXiv.
Generating diverse educational interfaces during learning design allows instructors to review and correct materials before they reach students, preventing unverified errors.
Rethinking Scaffolding In LLM Tutors: The Interactional Mismatch Between Benchmarks And Real-World Deployments
Alexandra Neagu, Jeffrey T. H. Wong, Marcus Messer, Rhodri Nelson, Peter B. Johnson. (06/2026). arXiv.
Students in real-world deployments showed substantially lower uptake of chatbot scaffolding than educational benchmarks implicitly assume.
"Stuck In A Spiral": Shame And Guilt As Social Regulators Of AI Use In Computing Education
Kate Hamilton, Irene Hou, Dev Patel, Sheena Nnam, Hena Patel, Stephen MacNeil. (06/2026). arXiv.
Shame and guilt regulated the visibility of AI use, leading students to hide their tool use through dimmed screens, switched tabs, and selective disclosure.
From Planning To Revision: How AI Writing Support At Different Stages Alters Ownership
Katy Ilonka Gero, Tao Long, Carly Schnitzler, Paramveer S. Dhillon. (06/2026). arXiv.
Any AI assistance lowered ownership compared to no AI, but drafting support caused the largest drop, while planning support preserved the most ownership.
Exploring How Agent Voice Accents Shape Human-AI Collaboration In K-12 Group Learning
Prerna Ravi, Carumey Stevens, Ben Hurt, Brandon Hanks, Grace Lin, and Emma Anderson. (06/2026). arXiv.
Surveys showed no statistically significant differences across agent accent conditions in teachers' ratings of proficiency, etiquette and mannerisms, or personality.
Large-Scale Semantic Mapping Of Learner Agency And Autonomy Reveals What Measurement And Generative AI Research Overlook
Fei Qin, Xiaobo Liu, Yaowen Zhang, Xuming Li, Fei Wang, Mutlu Cukurova, Jingjing Chen, Yu Zhang. (06/2026). arXiv.
Definitions spanned three latent dimensions—task-related learning regulation and control, intrinsic motivation and internal decisions, and social-relational action—rather than a binary division between agency and autonomy.
Intelicit: Eliciting And Assessing Contextualized Creativity Via Dialogue Policy Optimization
Mingjia Li, Jin Wu, Hong Qian, Wenhao Huang, Yiyang Huang, Yiwen Zhang, Chanjin Zheng, Xiangfeng Wang, and Aimin Zhou, Jiajun Guo. (06/2026). arXiv.
In simulation across 16 scenarios, IntElicit achieved a significantly higher average creativity score (325.71) than top models like GPT-4o (301.67) and DeepSeek-R1 (298.40).
Examining The Usage Of Generative AI Models In Student Learning Activities For Software Programming
Rufeng Chen, Shuaishuai Jiang, AJung Moon, Jiyun Shen, Lili Wei. (06/2026). arXiv.
Beginners with ChatGPT achieved significantly higher task scores (410.7 passed tests) than beginners without it (80.7), but generating complete solutions did not improve conceptual understanding.
Beyond Tool Adoption: A Practical Five-Stage Developmental Continuum For AI Literacy In Higher Education
Paul Liu, Rachel Levy. (06/2026). arXiv.
Brief one- to two-day training workshops enabled participants to demonstrate behaviors consistent with movement from non-engagement or uncritical use toward informed engagement.
