CS1-LLM: Integrating LLMs into CS1 Instruction

Overview

• Recent advancements in Large Language Models (LLMs) like ChatGPT and GitHub Copilot are impacting introductory programming courses (CS1)
• LLMs can now solve many of the assignments and exams previously used in CS1
• This raises questions about what should be taught and how to teach it, including whether students should be trained in using these tools
• This experience report describes a CS1 course that fully embraces the use of LLMs from the beginning

Plain English Explanation

The rapid development of powerful Large Language Models (LLMs) like ChatGPT and GitHub Copilot is changing the landscape of introductory programming courses (CS1) at universities. These advanced AI systems can now solve many of the same assignments and exams that CS1 students used to struggle with. This is prompting educators to rethink what they should be teaching and how to best prepare students for the new reality of widely available LLMs.

In this report, the authors describe a CS1 course that has fully embraced the use of LLMs from the very beginning. Instead of focusing heavily on syntax and writing code from scratch, the course now emphasizes the skills needed to successfully use an LLM to produce software. This includes explaining code, testing code, and breaking down larger problems into smaller, LLM-solvable tasks.

The course also features three large, open-ended projects in different domains like data science, image processing, and game design. These allow students to showcase their creativity and apply their LLM skills in practical, real-world scenarios. Student feedback suggests they appreciated learning with the assistance of the LLM and used it in a variety of ways when writing code.

The report provides lessons learned for instructors who may want to follow a similar approach and integrate LLMs into their own CS1 courses. It highlights the need to shift the focus away from low-level programming details and towards the higher-level skills of software development that will be crucial as LLMs become more ubiquitous.

Technical Explanation

The researchers describe a CS1 course at a large research university that has been intentionally redesigned to fully incorporate the use of Large Language Models (LLMs) from the beginning.

This course alteration was motivated by the recent research showing that LLMs can now solve the majority of assignments and exams that were previously used in CS1 courses. Additionally, professional software engineers are increasingly using these tools, prompting the question of whether students should also be trained in their use.

To embrace LLMs, the course reduced the emphasis on syntax and writing code from scratch. Instead, it now focuses on the skills needed to successfully produce software with an LLM, such as explaining code, testing code, and decomposing larger problems into smaller, LLM-solvable functions.

In addition to frequent, formative assessments of these skills, the course featured three large, open-ended projects in different domains (data science, image processing, and game design). These allowed students to showcase their creativity and apply their LLM skills in practical, real-world scenarios.

An end-of-term survey revealed that students appreciated learning with the assistance of the LLM and used it in a variety of ways when writing code. The report provides lessons learned for instructors who may wish to incorporate LLMs into their own CS1 courses, emphasizing the need to shift the focus away from low-level programming details and towards the higher-level skills of software development.

Critical Analysis

The experience report presents a compelling case for embracing the use of Large Language Models (LLMs) in introductory computer science education. By acknowledging the growing capabilities of these AI systems and adapting the course accordingly, the authors have demonstrated a proactive approach to preparing students for the evolving landscape of software development.

One potential limitation of the study is the lack of a direct comparison to a control group or traditional CS1 course. While the authors provide insights into student feedback and the overall approach, it would be valuable to have more quantitative data on the relative effectiveness of this LLM-focused curriculum compared to a more traditional one.

Additionally, the report does not delve into the potential challenges or drawbacks of over-relying on LLMs in an educational setting. For example, there may be concerns about students becoming overly dependent on the technology or failing to develop a deeper understanding of fundamental programming concepts. Addressing these potential issues could strengthen the research and provide a more well-rounded perspective.

Overall, the experience report offers a thoughtful and timely exploration of the role of LLMs in introductory computer science education. By sharing their lessons learned, the authors encourage other instructors to consider how they can adapt their own courses to best prepare students for the realities of the modern software development landscape.

Conclusion

This experience report highlights the significant impact that the widespread availability of Large Language Models (LLMs) is having on introductory programming courses (CS1) at the university level. By fully embracing the use of LLMs in a CS1 course, the researchers have demonstrated a proactive approach to adapting the curriculum and preparing students for the evolving nature of software development.

The key insights from this report include the need to shift the course focus away from low-level programming details and towards the higher-level skills required to effectively utilize LLMs, such as code explanation, testing, and problem decomposition. The inclusion of open-ended projects also allows students to showcase their creativity and apply their LLM skills in practical, real-world scenarios.

As LLMs continue to advance and become more widely adopted, this experience report provides a valuable blueprint for other instructors looking to integrate these powerful AI tools into their own CS1 courses. By sharing their lessons learned, the authors encourage a forward-thinking approach to computer science education that equips students with the necessary skills to thrive in an increasingly AI-driven world.