CodeSpectra the Illusion of Easy Coding: Why AI Still Demands Effort (blog.ptidej.net)

CodeSpectra, a newly developed plagiarism detection tool, addresses the growing challenges of academic integrity in coding amidst the rise of AI coding assistants like ChatGPT. By enabling students to quickly generate structurally similar code, traditional plagiarism checkers struggle to identify clever manipulations such as variable renaming or style alterations. CodeSpectra employs a sophisticated four-stage detection technique that meticulously analyzes submissions to ensure originality while providing educators with reliable reports to maintain academic standards. Each of the four stages is finely tuned for specific plagiarism types, enhancing performance without redundant processing. The stages begin with Exact-Match Hashing to catch direct clones, followed by Blind Normalization to address variable renaming, and then Machine Learning through RandomForest to detect structural similarities. The final stage examines semantic similarities, crucial for identifying submissions generated by AI. This multi-faceted approach not only scales efficiently for large submission cohorts but also reinforces the importance of educators’ judgments in a landscape increasingly influenced by automated solutions. As AI evolves, tools like CodeSpectra are essential for preserving integrity in the AI/ML learning environment.