Breadcrumb
Search and Filter
The Repository
Others
Research synthesis is AI-generated, human reviewed. Updated 11/2026.
Displaying 1 - 30 of 200
When Evaluators Cry Wolf: Lessons From Production LLM-As-Judge Evaluation In Educational AI
Chris Rohlfs, Rodrigo Vergara Bosse, Priscilla Rain Hopper, Keanon O'Keefe, Patrick Russell. (09/2026). arXiv.
Across 21 deployed evaluators, the calibration strategy cut confirmed false-positive flags by 99% and raised per-flag precision from 0.6% to 48.9%.
How University Disability Services Professionals Write Image Descriptions For Hci Figures Using Generative AI
Muhammad Raees, Yugo Iwamoto, Konstantinos Papangelis, Jamison Heard, Garreth W. Tigwell. (09/2026). arXiv.
HCI experts rated AI-assisted figure descriptions significantly higher than human-only descriptions across accuracy, completeness, clarity, effectiveness, and descriptiveness.
AI-Based Scoring Systematically Underestimates Conceptual Understanding Of Linguistically Weak Students' Explanations In Physics
Markus Sebastian Feser, Paul Tschisgale. (09/2026). arXiv.
Across all 11 AI scoring approaches, lower writing quality was associated with 1.3- to 3.9-times higher odds of underestimating students' conceptual understanding in physics.
LLM-As-A-Judge Validity Is Strongly Task-Dependent Across Physics Assessment Formats
Will Yeadon, Tom Hardy, Paul Mackay, and Elise Agra. (09/2026). arXiv.
AI models showed robust rank-order agreement with human marks on structured questions and code-based plots, but achieved near-zero rank-order agreement when marking short-form physics essays.
A Benchmark For LLM'S Understanding Of Middle School And High School Science Topics
Noah L. Schroeder, Yessy Eka Ambarwati, Yuji Zhang, ChengXiang Zhai. (09/2026). arXiv.
Most tested models scored above 90% accuracy on middle and high school standards, with larger model size not consistently predicting higher performance.
Characterizing LLM-Based Family Education Through The Lens Of Activity Theory: A Scoping Review Of The Hci Literature
Lan Luo, Yuqi Liang, Jie Cai, Anqi Wang, Dongyijie Primo Pan, Muzhi Zhou, Chun Yu, Pan Hui. (09/2026). arXiv.
Most reviewed studies focused on child–parent interaction (83%) and centered on language, narrative, AI literacy, and relational development rather than mathematical or scientific inquiry.
Evaluation Of Pre-Trained Models For Pedagogical Assessment Of Novel AI-Assisted Educational Questions
Michael Lawrence Castanares, Princess Ventures, Allan Tan. (09/2026). arXiv.
Traditional machine learning models and BERT suffered substantial drops in Bloom classification performance on out-of-distribution AI questions, while large language models maintained higher baseline accuracy.
"We'Ll Fix It Later": Education, AI, And The Deferral Of Privacy In EdTech
Meghna Manoj Nair, Rachel Greenstadt. (09/2026). arXiv.
One-third of audited platforms made no meaningful AI disclosure despite visible AI features, and 73% provided only generic accountability and breach-response language.
Biased AI Improves Human Performance But Reduces Perceived Helpfulness
Shiyang Lai, Jiwoong Choi, Junsol Kim, Nadav Kunievsky, Yujin Potter, James Evans. (09/2026). arXiv.
Biased AI assistants improved or preserved human performance relative to default and neutral AI across fact-checking, investment, and graduate coding tasks.
AI-Written Admissions Essays Are Widespread But Penalized
Calvin Isley, Johann Gaebler, Sharad Goel. (09/2026). arXiv.
By 2025, 56% of applicants submitted at least one primarily AI-generated essay—reaching 69.3% among international applicants—despite an explicit prohibition against AI use.
Human Resilience In The AI Era -- What Machines Can'T Replace
Shaoshan Liu, Anina Schwarzenbach, Yiyu Shi. (09/2026). arXiv.
Resilience operates across psychological, social, and organizational levels, combining individual adaptation with peer support and institutional recovery processes.
Finding Common Mistakes In Modelling With Mathematical Formalisms Using LLMs
Lilian Killich, Marko Schmellenkamp, Fabian Vehlken, Thomas Zeume. (09/2026). arXiv.
On a propositional logic data set of 6,106 student pairs, the workflow discovered transformation clusters explaining 84.44% of student mistakes, including 15.03% not captured by previous hand-picked analyses.
Limits Of LLM Text Detectors In Education
Lukas Gehring, Benjamin PaaBen. (09/2026). arXiv.
Detector performance drops sharply on texts with intermediate assistance, particularly AI-assisted rewrites of human-written essays.
Artificial Intelligence Literacy And Sustainable Development: An Ethical Governance And Development Goals Framework
Md. Masudul Islam, Mirza Niaz Morshed, Md. Shafiqul Islam. (09/2026). arXiv.
Governance literacy was the strongest predictor of sustainable development goal nexus awareness (standardized beta = 0.64), followed by ethical literacy and technical literacy.
Statistical Theory In The Age Of Machine-Assisted Mathematics: Rethinking How Theory Is Made And Taught
Pietro Coretto. (09/2026). arXiv.
Machine formalization identified 32 mathematical conditions across five statistical cases, linking each condition to specific proof steps that informal textbook citations usually compress.
The AI Adaptation Gap In Higher Education: Students, Faculty, And Administrative Staff
Yuriy S. Braun, Salavat M. Khafizov. (08/2026). arXiv.
Students reported higher AI-use intensity and perceived usefulness than faculty and staff, while faculty and staff reported stronger academic integrity concerns and responsible-use norms.
ChatGPT Solves All Tested Qiskit Homework Assignments
Alexei Kaltchenko, Gurnivaj Tiwana. (08/2026). arXiv.
ChatGPT produced final artifacts that executed and passed autograding in all 150 tested sessions across the three introductory quantum programming assignments.
Learnai: Just-In-Time AI Co-Creation Across Disciplines At A University
Weihao Qu, Ling Zheng, Chris Buzaid, Daniel Crawford. (08/2026). arXiv.
Clients across disciplines co-created 36 portfolio websites and over 20 deployed web applications with peer tutors.
Do Assessment Instruments Measure The Same Thing For Humans And LLMs? A Latent Structure Analysis
Alona Strugatski, Licol Zeinfeld, Giora Alexandron. (08/2026). arXiv.
Tests showed systematic differences between human and language model factor structures, indicating the exams may not measure the same constructs in humans and AI.
Probing AI-Generated Physics Solutions And Preparing Students To Critique Them
Nikhil Sanjay Borse, Amir Bralin, Sean Savage, N. Sanjay Rebello. (08/2026). arXiv.
Well-specified prompts produced complete quantitative solutions from o4-mini, whereas underspecified and multimodal prompts led to missing assumptions, skipped numerical integration, and misleading final results.
AI-Generated Interactive Fiction For Educational Use: A Pilot Study Of Perceived Comprehensibility, Coherence, And Engagement
F. Rogosch, A. Schrader. (08/2026). arXiv.
Participants rated story-content coherence as the weakest dimension (mean 2.92 out of 5), while rating narrative clarity (4.11) and length acceptance (4.14) positively.
Does Generative AI Narrow Education-Based Productivity Gaps? Evidence From A Randomized Experiment
Guillermo Cruces, Diego Fernandez Meijide, Sebastian Galiani, Ramiro Galvez, Maria Lombardi. (08/2026). arXiv.
AI assistance closed about three-quarters of the baseline performance gap between higher- and lower-education participants, raising scores for both groups but with substantially larger gains for lower-education adults.
When LLM Essays Outscore Student Essays: What A Korean Writing Rubric Rewards And Where Readers Disagree
Shinwoo Park, Yo-Sub Han. (08/2026). arXiv.
Large language model essays outscored student essays by an average of 18.35 points across a 100-point rubric, exceeding student averages across all four evaluated models.
What Remains Human In Mathematics In The Age Of AI
Amir Moradifam. (07/2026). arXiv.
As routine execution and computation become automated, distinctively human mathematical skills such as problem framing, conceptual insight, and creative judgment become more visible and valuable.
Agreement Is Not Quality: Blind Expert Verification Of Human And LLM Qualitative Coding When Human Consensus Is Not Ground Truth
Alex Liu, Lief Esbenshade, Michael Xiao, Victor Tian, Zachary Zhang, Kevin He, Min Sun. (07/2026). arXiv.
A blind expert preferred human and model coding at statistically indistinguishable rates (51.5% versus 48.5%), despite models having lower agreement with human coders than humans had with each other.
Four Years Of Genai: How Educators And Industry Adapted Their Assessment Strategies
May Mahmoud, Nisa Shahid, Izah Sohail, Gulshan Sharma, Hanan Salam, Sarah Nadi. (07/2026). arXiv.
Educators and hiring professionals agreed on the most important competencies, ranking critical evaluation of AI outputs, independent learning, and responsible tool use highest.
Generative AI Availability, Grades, And Student Satisfaction At A Large University
James M. Zumel Dumlao, Meng Wang, Zhonghan Xie, Junyao Hu, Ivan Bar, George Chaney III, Henry Gold, Misha Teplitskiy. (07/2026). arXiv.
Effects of ChatGPT's release on student grades were small and non-robust, ranging from 0.010 to 0.071 grade points depending on how pandemic trends were modeled.
What Does The Credential Still Certify? Cognitive Stewardship For AI-Mediated Education
Kai Yao. (07/2026). arXiv.
University guidance stated rules on permitted AI use more clearly than the evidence standards required to justify learning claims, scoring 2.47 versus 1.89 out of 4.
From Novelty To Normalisation: Tracking Changing Perceptions Of AI In Higher Education, 2024-2026
Juliana Gerard, Morgan Macleod, Kelly Norwood, Aisling Reid, Muskaan Singh. (07/2026). arXiv.
Familiarity and practical experience with AI tools increased across students and staff between 2024 and 2026, shifting AI from a novelty into routine academic practice.
Cost-Efficient Generative AI Summarization For Scalable Automated Essay Scoring In Educational Assessment
Haowei Hua. (07/2026). arXiv.
GPT-5 mini achieved the highest human-machine scoring agreement (QWK of 0.8435), slightly outperforming full GPT-5 (0.8350) and GPT-5 nano (0.8332).
