Fine-tuned decoder-only LLMs fall into a Semantic Trap on vulnerability detection, achieving high scores on unpaired normal code but failing on paired vulnerable-patched code, semantic perturbations, and gap analysis, while reasoning supervision reduces symptoms at the cost of recall.
Defecthunter: A novel llm-driven boosted-conformer-based code vulnerability detection mechanism,
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
years
2026 2verdicts
UNVERDICTED 2representative citing papers
JupOtter introduces notebook-specific tokenization, cell-level bug prediction, and OtterDataset to achieve higher F1 scores than static analyzers and LLMs on two of three evaluation datasets.
citing papers explorer
-
Do Fine-Tuned LLMs Understand Vulnerabilities? An Investigation into the Semantic Trap
Fine-tuned decoder-only LLMs fall into a Semantic Trap on vulnerability detection, achieving high scores on unpaired normal code but failing on paired vulnerable-patched code, semantic perturbations, and gap analysis, while reasoning supervision reduces symptoms at the cost of recall.
-
JupOtter: Cell-Level Bug Detection in Jupyter Notebooks
JupOtter introduces notebook-specific tokenization, cell-level bug prediction, and OtterDataset to achieve higher F1 scores than static analyzers and LLMs on two of three evaluation datasets.