To support novice learners of Java programming, we have developed the Java Programming Learning Assistant System (JPLAS). It offers several learning features, including the Code Writing Problem (CWP) where students need to write source codes that must pass the predefined JUnit test cases. However, plagiarism may occur where students share solutions or reuse codes, making it difficult to ensure the originality of submissions. In previous works, we presented a code plagiarism checking function using Levenshtein distance and a threshold selection method using Tukey's IQR-based, to detect potential plagiarism. We confirmed their validity through applications to source codes made by first-year undergraduate students in State Polytechnic of Malang (POLINEMA), Indonesia. In this study, we conduct this analysis to codes by 30 second-year undergraduate students and by generative AI across three CWP assignments. The results show that the average similarity score is higher for the second-year students than that for the first-year students, and the score is 100% for one code by the second-year student against an AI-generated one.