The scalability challenges of code review and pair-programming assessments in large computer science courses, such as CS50 at Harvard University, have opened up opportunities for the application of Generative AI. Leveraging large language models (LLMs), CS50.ai offers a suite of AI-based tools that assist both students and instructors in mastering course material while overcoming the limitations posed by human resource constraints. This demo highlights how generative AI can be employed to conduct code reviews and pair-programming simulations, providing real-time feedback, code explanations, and collaborative programming insights. By integrating these AI tools into students' learning journeys, we aim to mimic the 1:1 interaction between instructor and student, improving both formative and summative assessments. We will showcase how these tools are implemented to scale personalized feedback, ensure academic integrity, and maintain pedagogical efficacy. Our presentation will also reflect on lessons learned from deploying these AI-driven tools in recent course offerings.

In 2023, we developed and deployed AI-based tools in CS50 at Harvard University to provide students with 24/7 interactive assistance, approximating a 1:1 teacher-to-student ratio. These tools offer code explanations, style suggestions, and responses to course-related inquiries, emulating human educators to foster critical thinking. However, maintaining alignment with instructional goals is challenging, especially with frequent updates to the underlying large language models (LLMs). We thus propose a continuous improvement process for LLM-based systems using a collaborative human-in-the-loop approach. We introduce a systematic evaluation framework for assessing and refining the performance of AI-based tutors, combining human-graded and model-graded evaluations. Using few-shot prompting and fine-tuning, we aim to ensure our AI tools adopt pedagogically sound teaching styles. Fine-tuning with a small, high-quality dataset has shown significant improvements in aligning with teaching goals, as confirmed through multi-turn conversation evaluations. Additionally, our framework includes a model-evaluation backend that teaching assistants periodically review, ensuring the AI system remains effective and aligned with instructional objectives. This paper offers insights into our methods and the impact of these AI tools on CS50 and contributes to the discourse on AI in education, showcasing scalable, personalized learning enhancements.

In this paper, we present a study of the effects of structural holds and rigidity of a flexible display on touch pointing and dragging performance. We discuss an observational study in which we collected common holds used when pointing on a mockup paper display. We also measured the force patterns each hold generated within the display surface. We analyzed this data to produce 3 force zones in the display for each of the four most frequently observed holds: the grip zone, rigid zone, and the flexible zone. We report on an empirical evaluation in which we compared the efficiency of pointing and dragging operations between holds, and between structural zones within holds, using a real flexible Lumalive display. Results suggest that structural force distributions in a flexible display affect the Index of Performance of both pointing and dragging tasks, irrespective of hold, with rigid parts of the display yielding a 12% average performance gain over flexible areas.

In this paper, we introduce a multi-touch display surface that can be dynamically deformed from a flat circular display to a convex or concave hemispherical display. A rubber latex material is used for the display surface allowing it to inflate or deflate as air is pumped into or out of an airtight container. The elasticity of the display surface also allows users to deform the surface by varying the pressure exerted on it. This deformation may be detected by the device and provides a z-axis input in addition to the typical x,y-axis inputs of flat-screen displays.