How good is generative AI at passing exams? What does this tell us about how we could design better assessments? Join us on Monday 4th December at 2pm GMT (UTC) to discuss a paper on this by Joyce Mahon, Brian Mac Namee and Brett Becker at University College Dublin published at UKICER earlier this year.  From the abstract:
We investigate the capabilities of ChatGPT (GPT-4) on second level (high-school) computer science examinations: the UK A-Level and Irish Leaving Certificate. Both are national, government-set / approved, and centrally assessed examinations. We also evaluate performance differences in exams made publicly available before and after the ChatGPT knowledge cutoff date, and investigate what types of question ChatGPT struggles with.
We find that ChatGPT is capable of achieving very high marks on both exams and that the performance difference before and after the knowledge cutoff date are minimal. We also observe that ChatGPT struggles with questions involving symbols or images, which can be mitigated when in-text information ‘fills in the gaps’. Additionally, GPT-4 performance can be negatively impacted when an initial inaccurate answer leads to further inaccuracies in subsequent parts of the same question. Finally, the element of choice on the Leaving Certificate is a significant advantage in achieving a high grade. Notably, there are minimal occurrences of hallucinations in answers and few errors in solutions not involving images.
These results reveal several strengths and weaknesses of these exams in terms of how generative AI performs on them and have implications for exam design, the construction of marking schemes, and could also shift the focus of what is examined and how.
We’ll be joined by the papers lead author Joyce, who will give us a lightning talk summary of her paper to start our discussion. All welcome, as usual we’ll be meeting on zoom details at sigcse.cs.manchester.ac.uk/join-us
Joyce Mahon, Brian MacNamee and Brett A. Becker (2023) No More Pencils No More Books: Capabilities of Generative AI on Irish and UK Computer Science School Leaving Examinations. In The United Kingdom and Ireland Computing Education Research conference (UKICER 2023), September 07–08, 2023, Swansea, Wales UK. ACM, New York, NY, USA, 7 pages. DOI: 10.1145/3610969.3610982
Programming is hard, or at least it used to be. AI code generators like Amazon’s CodeWhisperer, DeepMind’s AlphaCode, GitHub’s CoPilot, Replit’s Ghostwriter and many others now make programming easier, at least for some people, some of the time. What opportunities and challenges do these new tools present for educators? Join us on Zoom to discuss an award winning paper by Brett Becker, Paul Denny, James Finnie-Ansley, Andrew Luxton-Reilly, James Prather and Eddie Antonio Santos at University College Dublin, the University of Auckland and Abilene Christian University on this very topic.  We’ll be joined by two of the co-authors who will present a lightning talk to kick-off our discussion, for our monthly ACM journal club meetup. Here’s the abstract of his paper:
The introductory programming sequence has been the focus of much research in computing education. The recent advent of several viable and freely-available AI-driven code generation tools present several immediate opportunities and challenges in this domain. In this position paper we argue that the community needs to act quickly in deciding what possible opportunities can and should be leveraged and how, while also working on overcoming otherwise mitigating the possible challenges. Assuming that the effectiveness and proliferation of these tools will continue to progress rapidly, without quick, deliberate, and concerted efforts, educators will lose advantage in helping shape what opportunities come to be, and what challenges will endure. With this paper we aim to seed this discussion within the computing education community.
Brett A. Becker, Paul Denny, James Finnie-Ansley, Andrew Luxton-Reilly, James Prather, Eddie Antonio Santos (2023) Programming Is Hard – Or at Least It Used to Be: Educational Opportunities and Challenges of AI Code Generation in Proceedings of the 54th ACM Technical Symposium on Computer Science Education: SIGCSE 2023, pages 500–506, DOI: 10.1145/3545945.3569759
Maybe you wrote that code and maybe you didn’t. If AI helped you, such as the OpenAI Codex in GitHub Copilot, how did it solve your problem? How much did Artificial Intelligence help or hinder your solution? Join us to discuss a paper by Michel Wermelinger from the Open University published in the SIGCSE technical symposium earlier this month on this very topic.  We’ll be joined by Michel who will present a lightning talk to kick-off our discussion. Here’s the abstract of his paper:
The teaching and assessment of introductory programming involves writing code that solves a problem described by text. Previous research found that OpenAI’s Codex, a natural language machine learning model trained on billions of lines of code, performs well on many programming problems, often generating correct and readable Python code. GitHub’s version of Codex, Copilot, is freely available to students. This raises pedagogic and academic integrity concerns. Educators need to know what Copilot is capable of, in order to adapt their teaching to AI-powered programming assistants. Previous research evaluated the most performant Codex model quantitatively, e.g. how many problems have at least one correct suggestion that passes all tests. Here I evaluate Copilot instead, to see if and how it differs from Codex, and look qualitatively at the generated suggestions, to understand the limitations of Copilot. I also report on the experience of using Copilot for other activities asked of students in programming courses: explaining code, generating tests and fixing bugs. The paper concludes with a discussion of the implications of the observed capabilities for the teaching of programming.
Michel Wermelinger (2023) Using GitHub Copilot to Solve Simple Programming Problems in Proceedings of the 54th ACM Technical Symposium on Computer Science Education Pages SIGCSE 2023 page 172–178 DOI: 10.1145/3545945.3569830