A carregar...
Publicação
Ai-driven software development - how has the rise of chatgpt impacted github users? Impact on code efficiency
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.55 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
This thesis analyses AI-driven Software Development practices, investigating how the ban of
ChatGPT in Italy affected GitHub activity using a Difference in Difference analysis. It covers
the process of retrieving and processing data from GitHub Archive to form a four-week dataset
spanning the period from 17/03/2023 to 14/04/2023. A scoring and selection approach to
identify users from Italy (Treatment Group) and Germany (Control Group) resulted in a dataset
comprising 244,401 commits from 10,520 individual GitHub users. Results suggest that users
affiliated with organizations show a 12.12% increase in GitHub events, implying a decrease in
coding efficiency after the ban of ChatGPT. In contrast, individual users' activities remain
largely unaffected by the ban. Coding errors rose by 8.91% on business days, further indicating
a reduction in code quality, while results for weekends and public holidays were insignificant.
Lastly, organization-related users active on business days exhibited a 13.39% increase in
GitHub events post-ban, suggesting a reduction in coding efficiency, and a 20.6% increase in
coding errors, pointing to a decline in code quality. No empirical evidence is found for the bans’
effects on collaboration practices. These findings suggest that ChatGPT is well integrated into
the daily software development workflow and actively used to assist in writing and debugging
code, especially in professional settings.
Descrição
Palavras-chave
Software development Chatgpt Github Code quality Difference in differences Llm