Teaching computer science at scale can be challenging. From our experience in CS50, Harvard University's introductory course, we've seen firsthand the impactful role that generative artificial intelligence can play in education. Recognizing its potential and stakes, we integrated OpenAI's GPT into our own teaching methodology. The goal was to emulate a 1:1 teacher-to-student ratio, incorporating ''pedagogical guardrails'' to maintain instructional integrity. The result was a personalized, AI-powered bot in the form of a friendly rubber duck aimed at delivering instructional responses and troubleshooting without giving outright solutions. In this tutorial, we share our journey and offer insights into responsibly harnessing AI in educational settings. Participants will gain hands-on experience working with GPT through OpenAI's latest APIs, understanding and crafting prompts, answering questions using embedding-based search, and, finally, collaboratively building their own AI chatbot. We will also explore the growing landscape of agentic AI tools such as Claude Code, GitHub CoPilot, OpenAI Codex, discussing their applications in educational contexts. Prior knowledge of Python is beneficial but not required, as all demo source code will be provided. Ultimately, we'll not only share lessons learned from our own approach but also equip educators hands-on with the knowledge and tools with which they, too, can implement these technologies in their unique teaching environments.

Micropuncture is a critical step in drug injection during retinal vein cannulation (RVC) surgery. Minimizing deformation during the micropuncture process is beneficial to reduce mechanical damage. However, this goal is challenging due to the viscoelastic characteristics of retinal tissue. In this paper, a robotic micropuncture scheme for deformation optimization that incorporates a nonlinear force model is proposed. Before micropuncture, a preload strategy is utilized to ensure stable contact between needle and retinal vein. Secondly, a nonlinear viscoelastic (NV) model is developed to characterize the nonlinearity and relaxation behavior of the tissue. Finally, a speed optimization framework, based on the NV model and physical constraint, is adopted to minimize deformation. The effectiveness of the proposed scheme is validated through in vitro experiments conducted on open-sky porcine eyes. With average force error of 1.48 μN, stable contact can be achieved via proportion-integral-differential controller. The experimental results demonstrate that the NV model is more suitable for force modeling of retinal tissue. Furthermore, the optimized speed results in an average deformation of 0.5727 mm, which represents a reduction of at least 21.02% compared to the linear model. Thanks to the proposed scheme, the robotic micropuncture based on a varying speed trajectory can reduce deformation and enhance the safety of RVC surgery.

Zebrafish are widely used in the biomedical field, as an ideal model for microinjection. In automated zebrafish microinjection, posture adjustment is the first and key step, which takes a lot of skill, and injection success assessment is a challenging task. Constrained by these two aspects, it is difficult to further enhance the efficiency and success rate of injection. In this study, we propose an automated dual-micropipette coordination microinjection system. Zebrafish are randomly arranged in our system, reducing the operational difficulty, and the yolk is positioned using a pose estimation algorithm, followed by injection accomplished with dual-micropipette. Due to the reduction of posture adjustment time by half, the proposed system achieves the shortest injection time of 15.2s. Moreover, the simplicity of the system and the ease of operation contribute to the clinical feasibility of our system.

Teaching computer science at scale can be challenging. From our experience in CS50, Harvard University's introductory course, we've seen firsthand the impactful role that generative artificial intelligence can play in education. Recognizing its potential and stakes, we integrated OpenAI's GPT into our own teaching methodology. The goal was to emulate a 1:1 teacher-to-student ratio, incorporating "pedagogical guardrails" to maintain instructional integrity. The result was a personalized, AI-powered bot in the form of a friendly rubber duck aimed at delivering instructional responses and troubleshooting without giving outright solutions. In this tutorial, we share our journey and offer insights into responsibly harnessing AI in educational settings. Participants will gain hands-on experience working with GPT through OpenAI's latest APIs, understanding and crafting prompts, answering questions using embedding-based search, and finally, collaboratively building their own AI chatbot. Ultimately, we'll not only share lessons learned from our own approach but also equip educators hands-on with the knowledge and tools with which they, too, can implement these technologies in their unique teaching environments.

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.

Retinal vein occlusion (RVO) is a prevalent ocular pathology that may result in hemorrhage and even blindness. Currently, a procedure termed retinal vein cannulation (RVC), involving puncturing the retinal vein and injecting medication, has been developed. However, RVC requires extremely high precision at the micron scale. To address the challenges in the micro-puncturing process of the RVC, an adaptive finite-time sliding mode (AFSM) control scheme with a smooth motion generator has been proposed to assist surgeons in achieving precise micro-punctures using a piezo-actuated stage. Firstly, an S-curve-based smooth motion planning approach incorporating force feedback is designed to detect the successful micro-puncture state, thus addressing the challenge of limited force perception during the procedure. Subsequently, an AFSM control scheme has been developed to track the desired motion. Finally, a micro-puncture system, equipped with a silicon phantom, is established for experimental purposes. The experimental results demonstrate that the proposed control scheme significantly enhances the tracking performance during the micro-puncture process. The smooth motion planning and AFSM control scheme prove to be effective for the automatic control of the piezo-actuated end-effector, thereby providing improved assistance to surgeons in the RVC process. Note to Practitioners—During the procedure of retinal vein micro-puncture, it is crucial to ensure a smooth motion planning and accurate tracking to guide the needle tip into the retinal vein lumen. In light of the difficulties in lack of depth perception, a motion generator has been proposed with an adaptive micro-puncture state detection mechanism based on force feedback. To reliably track the desired motion, the AFSM controller is designed to ensure tracking accuracy and robustness, and finite-time stability. In particular, the adaptive gain of the AFSM controller does not require uncertain prior information, making it friendly to clinical applications. The experimental results based on silicone phantom, demonstrate the effectiveness of the controller in achieving successful micro-puncture with precise tracking performance. The implementation of the AFSM controller enables the automated micro-puncture task, reducing the risk of damage during operation.

CS50.dev is a cloud-based programming environment offered to students taking CS50 and other CS courses at Harvard University, both on-campus or online. Built atop GitHub Codespaces, CS50.dev simplifies the initial challenges commonly faced by students and instructors because of the complexities involved in setting up programming environments at term's start. This demo offers an in-depth exploration of CS50.dev's architecture and presents a detailed guide on customizing Docker images and development container (devcontainers) to meet the specific needs of courses within GitHub Codespaces. The demo will also provide general guidance on how to help students transition from CS50.dev to using VS Code independently on their local machines at the term's end.

Teaching computer science at scale can be challenging. From our experience in CS50, Harvard University's introductory course, we've seen firsthand the impactful role that generative artificial intelligence can play in education. Recognizing its potential and stakes, we integrated OpenAI's GPT into our own teaching methodology. The goal was to emulate a 1:1 teacher-to-student ratio, incorporating "pedagogical guardrails" to maintain instructional integrity. The result was a personalized, AI-powered bot in the form of a friendly rubber duck aimed at delivering instructional responses and troubleshooting without giving outright solutions. We plan to share our journey and offer insights into responsibly harnessing AI in educational settings. Participants will gain hands-on experience working with GPT through OpenAI's APIs, understanding and crafting prompts, answering questions using embedding-based search, and finally, building their own AI chatbot. Ultimately, we'll not only share lessons learned from our own approach but also equip educators hands-on with the knowledge and tools with which they, too, can implement these technologies in their unique teaching environments.

In Summer 2023, we developed and integrated a suite of AI-based software tools into CS50 at Harvard University. These tools were initially available to approximately 70 summer students, then to thousands of students online, and finally to several hundred on campus during Fall 2023. Per the course's own policy, we encouraged students to use these course-specific tools and limited the use of commercial AI software such as ChatGPT, GitHub Copilot, and the new Bing. Our goal was to approximate a 1:1 teacher-to-student ratio through software, thereby equipping students with a pedagogically-minded subject-matter expert by their side at all times, designed to guide students toward solutions rather than offer them outright. The tools were received positively by students, who noted that they felt like they had "a personal tutor.'' Our findings suggest that integrating AI thoughtfully into educational settings enhances the learning experience by providing continuous, customized support and enabling human educators to address more complex pedagogical issues. In this paper, we detail how AI tools have augmented teaching and learning in CS50, specifically in explaining code snippets, improving code style, and accurately responding to curricular and administrative queries on the course's discussion forum. Additionally, we present our methodological approach, implementation details, and guidance for those considering using these tools or AI generally in education.

CS50.ai is an AI-based educational tool developed and integrated into CS50 at Harvard University using large language models (LLMs), supporting both in-person and online learners. CS50.ai encapsulates a variety of AI-based tools designed to enhance students' learning by approximating a 1:1 teacher-to-student ratio. We showcase: "Explain Highlighted Code," a Visual Studio (VS) Code extension that provides just-in-time explanations of code snippets; style50, a VS Code extension that offers formatting suggestions and explanations thereof; and our "CS50 Duck," an AI-based chatbot for course-related questions, implemented both as a VS Code extension and as a standalone web application. We also demonstrate the integration of our tools into Ed, the course's discussion forum. This demo will illustrate the functionality and effectiveness of these tools as well as the pedagogical "guardrails" that we put in place to ensure secure and fair usage of these tools, while sharing insights from our own experience therewith this past summer and fall.

For CS50 at Harvard, we have long provided students with a standardized programming environment, to avoid start-of-term technical difficulties that might otherwise arise if students had to install and configure compilers, interpreters, and debuggers on their own Macs and PCs. (For many students, "hello, world" is challenge enough on day 0, without also encountering "command not found" at the same time!) We originally provided students with shell accounts on a university-managed cluster of systems. We then transitioned to a cloud-based equivalent so as to manage the systems ourselves, root access and all. We transitioned thereafter to client-side virtual machines, to scale to more students and enable GUI-based assignments. We have since transitioned to web-based environments, complete with code tabs, terminal windows, and file explorers, initially implemented atop AWS Cloud9 and now, most recently, GitHub Codespaces, an implementation of Visual Studio (VS) Code in the cloud, free for teachers and students alike. In this workshop, we'll discuss the pedagogical and technological advantages and disadvantages of every approach and focus most of our time, hands-on, on using and configuring GitHub Codespaces itself for teaching and learning. Along the way, attendees will learn how to create their own Docker images and "devcontainers" for their own classes and any languages they teach. Attendees will learn what is possible educationally by writing their own VS Code extensions as well. And how, at term's end, to "offboard" students to VS Code itself on their own Macs and PCs, so as to continue programming independent of Codespaces.

Zebrafish microinjection is widely used in vascular biology, neurology, and other research areas. The microinjection of zebrafish larvae is a complicated 3D task because zebrafish larvae are independent individuals with complex body structures compared with biological cells. In this study, we propose an automatic high-throughput injection system for zebrafish larvae under an optical inverted microscope. The proposed system combines the fixing device design, object detection, visual positioning, and 3D injection path planning to improve the survival rate of zebrafish after injection. Experimental results demonstrated the capability and efficiency of the proposed system, which achieved a success rate of 95.0% and a survival rate of 98.7% for zebrafish injection. The system can be applied to various biomedical and biochemical experiments.Note to Practitioners—Zebrafish larvae have long been an important model organism in neuroscience, biomedicine, and drug discovery. In various methods for external substance transportation into the zebrafish, microinjection is more efficient but also more difficult. At present, high-throughput injection of zebrafish larvae is mainly based on traditional manual operations. The injection speed and success rate of manual injection gradually decrease with the increase in zebrafish numbers. In this study, an automatic injection system is designed and implemented for high-throughput zebrafish larvae injection. We first use a simply designed mold to fix batch zebrafish larvae. Then we propose a coarse-fine two-step positioning method and perform a 3D injection path planning to achieve precise microinjection. The proposed injection system frees the operators from the repetitive and tedious injection work, providing technical support for the life science experiment.

This workshop introduces participants hands-on to GitHub Classroom, a free, web-based application that enables teachers to distribute, collect, and autograde programming assignments in any language. GitHub Classroom effectively provides an abstraction layer atop Git and GitHub, automating tasks for students and teachers that would otherwise be time-consuming and tedious via a CLI or GUI, particularly for larger courses. To distribute an assignment, a teacher need only create a "repository" with starter code (and optional tests) in their own "organization" on GitHub, which students can then "accept," automatically creating within that same organization a copy thereof to which the student has write access. To submit work, students need only upload (or push) files to the same, which automates execution of any tests. Students' submissions (and the results of any such tests) are then accessible via the teacher's own dashboard. GitHub Classroom does not strictly require familiarity with Git of students and teachers; basic operations can be performed via GitHub's GUI. But some comfort with cloning, adding, committing, and pushing files with Git itself is ideal. By workshop's end, participants will be prepared to use GitHub Classroom in their own courses if they so choose. And by way of GitHub Classroom can participants' students acquire all the more comfort with Git itself by courses' end.

For CS50 at Harvard, we have long provided students with a standardized programming environment, to avoid start-of-term technical difficulties that might otherwise arise if students had to install and configure compilers, interpreters, and debuggers on their own Macs and PCs. (For many students, "hello, world" is challenge enough on day 0, without also encountering "command not found" at the same time!) We originally provided students with shell accounts on a university-managed cluster of systems. We then transitioned to a cloud-based equivalent so as to manage the systems ourselves, root access and all. We transitioned thereafter to client-side virtual machines, to scale to more students and enable GUI-based assignments. We have since transitioned to web-based environments, complete with code tabs, terminal windows, and file explorers, initially implemented atop AWS Cloud9 and now, most recently, GitHub Codespaces, an implementation of Visual Studio (VS) Code in the cloud, free for teachers and students alike. In this workshop, we'll discuss the pedagogical and technological advantages and disadvantages of every approach and focus most of our time, hands-on, on using and configuring GitHub Codespaces itself for teaching and learning. Along the way, attendees will learn how to create their own Docker images and "devcontainers" for their own classes and any languages they teach. Attendees will learn what is possible educationally by writing their own VS Code extensions as well. And how, at term's end, to "offboard" students to VS Code itself on their own Macs and PCs, so as to continue programming independent of Codespaces.