Enhancing Learning via AI-Generated Feedback and Resubmission of Formative Assessments

Large Language Models (LLMs) are equipped with strong natural language understanding abilities. These abilities can be leveraged to evaluate and provide feedback on student work, provided that the model is properly prompted. We find that this feedback is more detailed, personalized, and requires fewer resources to produce than human-generated feedback. Due to the reduction in resources, students were permitted to resubmit their formative assessments throughout a course. Results show that students who consistently resubmitted performed statistically significantly better on their summative assessments, despite having indistinguishable patterns of access to course materials.