mcdok at SemEval-2026 Task 13: Finetuning LLMs for Detection of Machine-Generated Code

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, short

Summary

Adam Skurla, Dominik Macko, and Jakub Simko presented their "mcdok" system at SemEval-2026 Task 13, focusing on multi-domain detection of machine-generated code snippets across various programming languages. This task addresses challenges including binary detection, source attribution, identifying the generator LLM family, and detecting hybrid or adversarially modified code. The "mcdok" approach, originally designed for machine-generated text, was adapted for code understanding by exploring different base models. Their systems demonstrated competitive performance across all three subtasks. However, the authors noted significant performance gaps compared to the top-performing systems, indicating substantial room for further development and refinement in this area. The work was published in the Proceedings of the 20th International Workshop on Semantic Evaluation (2026), pages 322-327.

Key takeaway

For Machine Learning Engineers developing systems to detect machine-generated code, you should recognize that finetuned LLMs offer competitive performance but still face significant challenges. Your current approaches may struggle with hybrid human-machine code or adversarially modified snippets. Consider focusing your research and development on more robust finetuning strategies or novel architectural designs to bridge the substantial performance gap observed in SemEval-2026 Task 13, especially for complex detection scenarios.

Key insights

Finetuning LLMs for code understanding enables multi-domain detection of machine-generated code, including hybrid and adversarial forms.

Principles

• Machine-generated text detection approaches can be adapted for code.
• Code generation detection requires handling diverse languages and origins.

Method

The "mdok" approach, originally for text, was adjusted for code by exploring and finetuning various base models optimized for code understanding.

Topics

• Machine-Generated Code Detection
• Large Language Models
• LLM Finetuning
• Code Understanding
• SemEval-2026
• Code Attribution

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• mdok-style at SemEval-2026 Task 10: Finetuning LLMs for Conspiracy Detection — Data augmentation and self-training can effectively finetune LLMs for specialized text classification tasks with limited data.
• Beyond Problem Solving: UOJ-Bench for Evaluating Code Generation, Hacking, and Repair in Competitive Programming — LLMs can complement traditional online judges by identifying subtle code errors, but effective error detection remains computationally intensive.
• A NATO-accredited organization published a report concerning language-based disparities in LLM performance (Published April 25, 2026) — LLM performance varies significantly by language, but adaptation strategies can bridge the gap.
• Calibration Without Comprehension: Diagnosing the Limits of Fine-Tuning LLMs for Vulnerability Detection in Systems Software — LLMs fine-tuned for vulnerability detection achieve "calibration without comprehension," adapting outputs without true security reasoning.
• LLMs Made Simple: Examples, Analogies & Memory Tricks — LLMs are advanced autocomplete systems predicting tokens to generate human-like language from vast training data.
• Contrastive Training with LLM-generated Near-Misses for Robust Code-Switching Speech Recognition — Contrastive training with LLM-generated near-misses robustly improves code-switching ASR at critical language-alternation points.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.