ETD

• Chain of Thoughts
• Large Language Models

Large Language Models have proven effective in analyzing, understanding, and generating code. Recently, the findings on reasoning topologies and structured pipelines have significantly enhanced model performance, providing a rich framework for designing specialized methodologies for solving specific tasks. This thesis introduces Chain of Concurrent Static Analysis, a specialized reasoning chain designed to detect and classify concurrent bugs in Go source code. The proposed framework was evaluated on a dataset of 103 buggy code snippets, and its performance was compared against baseline I/O prompting and models that support chain of thoughts reasoning; showing an overall increase in vulnerability detection and comprehension.