Szczegóły publikacji
Opis bibliograficzny
Investigating code similarity patterns in LLM-generated and human-written programming solutions / Paulina GACEK, Bartosz Gdowski, Konrad Szymański, Wojciech Żmuda // W: Proceedings of the 18th International Conference on Computer Supported Education [Dokument elektroniczny] : May 18–20, 2026, Benidorm, Spain , Vol. 1 / eds. Edmundo Tovar, Tania Di Mascio, Christoph Meinel. — Wersja do Windows. — Dane tekstowe. — [Spain] : ScitePress, [2026]. — ( CSEDU ; ISSN 2184-5026 ). — e-ISBN: 978-989-758-833-4. — S. 478–485. — Wymagania systemowe: Adobe Reader. — Tryb dostępu: https://www.scitepress.org/PublicationsDetail.aspx?ID=pYkGcjH... [2026-09-01]. — Bibliogr. s. 485, Abstr. — Dostęp po zalogowaniu
Autorzy (4)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 169730 |
|---|---|
| Data dodania do BaDAP | 2026-09-02 |
| DOI | 10.5220/0014949400004021 |
| Rok publikacji | 2026 |
| Typ publikacji | materiały konferencyjne (aut.) |
| Otwarty dostęp |  |
| Creative Commons | |
| Konferencja | International Conference on Computer Supported Education 2026 |
Abstract
As Large Language Models (LLMs) reshape programming education, detecting AI-generated submissions using traditional similarity-based plagiarism tools presents a novel challenge. This paper investigates whether LLM-generated code exhibits structural and lexical patterns that systematically differ from human-written solutions. Analyzing a dataset of 9,000 solutions to constrained algorithmic tasks, we compare pre-2023 human submissions against synthetic outputs from GPT-5.2 and Gemini 3.0 Flash. Our findings reveal that AI solutions occupy a highly concentrated space, exhibiting extreme structural convergence (¿75% similarity compared to 28% for humans) and highly constrained lexical overlap. While commenting behaviors vary significantly by model, these overarching convergences suggest similarity-based detection is theoretically viable. However, due to severe risks of model-specific statistical masking and false positives, strict pedagogical precautions are required before deployment.
