Inim- ja tehisaru võrdlus kursuse „Veebilehtede loomine edasijõudnutele“ lõpuprojektide näitel

Abstract

This bachelor's thesis aims to analyze the websites created as final projects of the University of Tartu course “Web Page Creation for Advanced Users” and to determine whether different language models are suitable for their evaluation. The evaluation was based on the criteria, made up of scientific literature. The theoretical part describes the course build, the importance of user interface, user experience, code writing in web development and gives an overview of artificial intelligence. The first step of the practical part gives the author's evaluation of the created prototypes of websites and the code of the final projects. While user interface and user experience can be considered satisfied, the quality of code is a bit more of a concern. The second step includes an investigation of different language models and their capabilities as of 20.02.2025 to evaluate the design based on screenshots. It turned out that DeepSeek and Copilot are not suitable for further investigation because of no possibility in evaluating screenshots. The final step of the practical part considerscomparison of author and language models evaluations. The study results showed that evaluations by Gemini 2.0 Flash and Gemini 2.5 Pro models did not significantly differ from the author’s ones. However, if the evaluation focused only on the user interface design, results by ChatGPT 4.5 and ChatGPT 4o, in addition to previously mentioned language models, did not significantly differ from the author's results. In the usability analysis, the ratings given by most language models, except for ChatGPT 4.5, statistically did not differ from the author ones. The code quality evaluations by language models varied.