The paper "Enhancing LLM-Based Feedback: Insights from Intelligent Tutoring Systems and the Learning Sciences" by John Stamper, Ruiwei Xiao, and Xinying Hou explores the integration of Large Language Models (LLMs) into Intelligent Tutoring Systems (ITSs) for generating feedback. The authors emphasize the importance of grounding LLM-based feedback generation in theoretical frameworks and empirical evidence to ensure its effectiveness in educational settings. They review existing research on feedback generation in ITSs, highlighting three primary methods: expert-created learner models, data-driven learner models, and the use of LLMs. The paper discusses the strengths and limitations of each method, particularly focusing on the role of LLMs in generating adaptive and human-like feedback. The authors advocate for a more cautious and theoretically grounded approach to LLM-based feedback, suggesting that current practices often lack a solid theoretical foundation and empirical validation. They propose a strategic blueprint for integrating LLMs into ITSs, emphasizing the need for careful design and evaluation of feedback content, trigger mechanisms, and delivery modalities. The paper also highlights the importance of incorporating learning sciences principles, such as Bloom’s Taxonomy and the Knowledge-Learning-Instruction (KLI) framework, to enhance the quality and effectiveness of feedback. Key recommendations include: 1. **Trigger Mechanisms**: Designing feedback triggers that are appropriate and timely, considering both student needs and potential over-reliance on LLMs. 2. **Prompt Engineering**: Using guidelines like the CLEAR framework to create effective prompts for LLMs, incorporating various types of information such as student models and learning objectives. 3. **Content Selection**: Optimizing feedback content based on learning sciences frameworks to promote long-term learning, such as using spaced repetition, testing, and faded worked examples. 4. **Delivery Modalities**: Expanding feedback delivery beyond text to include images, audios, and videos to enhance engagement and understanding. 5. **Evaluation**: Conducting comprehensive evaluations that include system performance, expert assessments, and classroom deployment data to ensure feedback quality and address ethical concerns. The paper concludes by emphasizing the importance of integrating generative AI with pedagogical design to fully leverage its potential in education, while maintaining a focus on evidence-based and theoretically grounded approaches.The paper "Enhancing LLM-Based Feedback: Insights from Intelligent Tutoring Systems and the Learning Sciences" by John Stamper, Ruiwei Xiao, and Xinying Hou explores the integration of Large Language Models (LLMs) into Intelligent Tutoring Systems (ITSs) for generating feedback. The authors emphasize the importance of grounding LLM-based feedback generation in theoretical frameworks and empirical evidence to ensure its effectiveness in educational settings. They review existing research on feedback generation in ITSs, highlighting three primary methods: expert-created learner models, data-driven learner models, and the use of LLMs. The paper discusses the strengths and limitations of each method, particularly focusing on the role of LLMs in generating adaptive and human-like feedback. The authors advocate for a more cautious and theoretically grounded approach to LLM-based feedback, suggesting that current practices often lack a solid theoretical foundation and empirical validation. They propose a strategic blueprint for integrating LLMs into ITSs, emphasizing the need for careful design and evaluation of feedback content, trigger mechanisms, and delivery modalities. The paper also highlights the importance of incorporating learning sciences principles, such as Bloom’s Taxonomy and the Knowledge-Learning-Instruction (KLI) framework, to enhance the quality and effectiveness of feedback. Key recommendations include: 1. **Trigger Mechanisms**: Designing feedback triggers that are appropriate and timely, considering both student needs and potential over-reliance on LLMs. 2. **Prompt Engineering**: Using guidelines like the CLEAR framework to create effective prompts for LLMs, incorporating various types of information such as student models and learning objectives. 3. **Content Selection**: Optimizing feedback content based on learning sciences frameworks to promote long-term learning, such as using spaced repetition, testing, and faded worked examples. 4. **Delivery Modalities**: Expanding feedback delivery beyond text to include images, audios, and videos to enhance engagement and understanding. 5. **Evaluation**: Conducting comprehensive evaluations that include system performance, expert assessments, and classroom deployment data to ensure feedback quality and address ethical concerns. The paper concludes by emphasizing the importance of integrating generative AI with pedagogical design to fully leverage its potential in education, while maintaining a focus on evidence-based and theoretically grounded approaches.