Breadcrumb
- Home
- AI Hub For Education
- Research Study Repository
- Evidence Explorer
- AI Tools
- Other AI Model
The Repository
Search and Filter
Other AI model
Studies tagged with AI from their full text or abstract; see each study for details.
Showing 391 - 412 of 412 studies
Fine-tuning ChatGPT for Automatic Scoring
Ehsan Latif, Xiaoming Zhai. (12/2023). arXiv.
Fine-tuned GPT-3.5-turbo achieved an average scoring accuracy of 91.5%, significantly outperforming fine-tuned BERT by 9.1 percentage points across six science assessment tasks.
Revolutionizing Customer Interactions: Insights and Challenges in Deploying ChatGPT and Generative Chatbots for FAQs
Feriel Khennouche, Youssef Elmira, Nabil Djebari, Yassine Himeur, Abbes Amira. (11/2023). arXiv.
Chatbot approaches for FAQs fall into four distinct categories: rule-based, retrieval-based, generative models, and hybrid architectures that combine retrieval with generative fallback.
The NCTE Transcripts: A Dataset of Elementary Math Classroom Transcripts
Dorottya Demszky, Heather Hill. (10/2023). arXiv.
Machine-identified dialogic moves—teacher uptake of student ideas, focusing questions, and student reasoning—are positively and significantly correlated with teachers' contributions to student test score gains.
Automating question generation from educational text
Ayan Kumar Bhowmick, Ashish Jagmohan, Aditya Vempaty, Prasenjit Dey, Leigh Hall, Jeremy Hartman, Ravi Kokku, Hema Maheshwari. (09/2023). arXiv.
In human evaluations, 93.5% of questions generated by the fixed prompt InstructGPT model and 92% from the hybrid model were deemed desirable quality.
ArguGPT: evaluating, understanding and identifying argumentative essays generated by GPT models
Yikang Liu, Ziyin Zhang, Wanyang Zhang, Shisen Yue, Xiaojing Zhao, Xinyuan Cheng, Yiwen Zhang, Hai Hu. (09/2023). arXiv.
English instructors distinguished human from GPT essays with 61.6% accuracy initially, which rose to 67.7% after one round of minimal self-training.
Designing a Realistic Peer-like Embodied Conversational Agent for Supporting Children's Storytelling
Zhixin Li, Ying Xu. (09/2023). arXiv.
The proposed agent, STARie, combines large language models, real-time voice cloning, and 3D facial animation to simulate an 8-year-old storytelling partner on a tablet interface.
Inspecting Spoken Language Understanding from Kids for Basic Math Learning at Home
Eda Okur, Saurav Sahay, Roddy Fuentes Alba, Lama Nachman. (06/2023). arXiv.
Whisper-medium achieved the lowest average word error rate (0.384 after cleaning and normalization), outperforming Google Cloud and Rockhopper speech recognition systems.
SIGHT: A Large Annotated Dataset on Student Insights Gathered from Higher Education Transcripts
Rose Wang, Pawan Wirawarn, Noah Goodman, Dorottya Demszky. (06/2023). arXiv.
Human-model agreement mirrored human consistency: categories with over 0.9 inter-rater reliability had agreement over 0.7, while less consistent categories (0.7-0.8) yielded lower agreement (0.3-0.5).
Interpretable Math Word Problem Solution Generation Via Step-by-step Planning
Mengxue Zhang, Zichao Wang, Zhichao Yang, Weiqi Feng, Andrew Lan. (06/2023). arXiv.
Planning-LM increased math problem solve rates from 8.1% to 14.1% for GPT-2 (117M) and achieved 72.3% on GPT-3, outperforming chain-of-thought prompting.
Computing Education in the Era of Generative AI
Paul Denny, James Prather, Brett A. Becker, James Finnie-Ansley, Arto Hellas, Juho Leinonen, Andrew Luxton-Reilly, Brent N. Reeves, Eddie Antonio Santos, Sami Sarsa. (06/2023). arXiv.
OpenAI's Codex scored 78.5% and 78.0% on two CS1 exams, ranking 17th out of 71 students in the class's top quartile.
The BEA 2023 Shared Task on Generating AI Teacher Responses in Educational Dialogues
Anaïs Tack, Ekaterina Kochmar, Zheng Yuan, Serge Bibauw, Chris Piech. (06/2023). arXiv.
The top system, built on GPT-3.5 with prompt ensembling and reranking, ranked first in both automated metrics and human evaluations of pedagogical ability.
Towards Applying Powerful Large AI Models in Classroom Teaching: Opportunities, Challenges and Prospects
Kehui Tan, Tianqi Pang, Chenyou Fan, Song Yu. (06/2023). arXiv.
Typical 40-minute classroom sessions generate over 12,000 tokens across 120 to 140 turns, exceeding the standard 4,096-token input limits of models like GPT-3.5.
Scalable Educational Question Generation with Pre-trained Language Models
Sahan Bulathwela, Hamze Muse, Emine Yilmaz. (05/2023). arXiv.
Pre-training the T5 model on scientific text abstracts significantly improved question predictive performance on the SciQ benchmark over the baseline model.
Let GPT be a Math Tutor: Teaching Math Word Problem Solvers with Customized Exercise Generation
Zhenwen Liang, Wenhao Yu, Tanmay Rajpurohit, Peter Clark, Xiangliang Zhang, Ashwin Kaylan. (05/2023). arXiv.
Student models trained with customized exercises achieved 94.7% accuracy on MAWPS and 93.3% on ASDiv-a, outperforming previous fine-tuned baselines and large language models.
The Utility of Large Language Models and Generative AI for Education Research
Andrew Katz, Umair Shakir, Benjamin Chambers. (05/2023). arXiv.
Sentence embeddings and GPT-3.5 successfully grouped 9,105 essay sentences into hierarchical clusters and concise summaries, creating a bottom-up codebook of student career interests.
Understanding EFL Students' Idea Generation Strategies for Creative Writing with NLG Tools
David James Woo, Yanzhi Wang, Hengky Susanto, Kai Guo. (04/2023). arXiv.
Students searched for ideas using language-level features such as words or sentences rather than broader discourse elements like plot or character.
ChatGPT or academic scientist? Distinguishing authorship with over 99% accuracy using off-the-shelf machine learning tools.
Heather Desaire, Aleesa E. Chua, Madeline Isom, Romana Jarosova, David Hua. (03/2023). arXiv.
The 20-feature model distinguished human and ChatGPT writing with 99.5% accuracy at the document level in training and 100% across two test sets.
Logic Against Bias: Textual Entailment Mitigates Stereotypical Sentence Reasoning
Hongyin Luo, James Glass. (03/2023). arXiv.
Textual entailment models substantially outperformed baseline similarity encoders on overall bias-fairness association scores across stereotype, profession, and emotion tests without explicit debiasing.
Build-a-Bot: Teaching Conversational AI Using a Transformer-Based Intent Recognition and Question Answering Architecture
Kate Pearce, Sharifa Alghowinem, Cynthia Breazeal. (12/2022). arXiv.
Build-a-Bot establishes a multi-step language pipeline that teaches students data collection, backtranslation augmentation, rule filtering, intent recognition, and question answering.
Solving math word problems with process- and outcome-based feedback
Jonathan Uesato, Nate Kushman, Ramana Kumar, Francis Song, Noah Siegel, Lisa Wang, Antonia Creswell, Geoffrey Irving, Irina Higgins. (11/2022). arXiv.
Combining supervised learning with reward-model-based reinforcement learning reduced GSM8K final-answer errors to 12.7% and reasoning trace errors to 3.4%.
The AI Teacher Test: Measuring the Pedagogical Ability of Blender and GPT-3 in Educational Dialogues
Anaïs Tack, Chris Piech. (05/2022). arXiv.
Conversational agents were quantifiably worse than real teachers across three pedagogical dimensions, falling furthest behind in helpfulness.
StoryBuddy: A Human-AI Collaborative Chatbot for Parent-Child Interactive Storytelling with Flexible Parental Involvement
Zheng Zhang, Ying Xu, Yanhao Wang, Bingsheng Yao, Daniel Ritchie, Tongshuang Wu, Mo Yu, Dakuo Wang, Toby Jia-Jun Li. (03/2022). arXiv.
All 12 parent-child pairs successfully used StoryBuddy to complete interactive storytelling sessions in both parent-assisted and automated bot-reading modes.
