Poster: gptCombFuzz: Combinatorial Oriented LLM Seed Generation for effective Fuzzing
Darshan Lohiya, Monika Rani Golla, Sangharatna Godboley, Pisipati Radha Krishna · 2024
The important contribution that large language models (LLMs) have made to the development of a new software testing era is the main objective of this proposed approach. It emphasizes the role that LLMs play in producing complex and diverse input seeds, which opens the way for efficient bug discovery. In the study we also introduce a systematic approach for combining various input values, employing the principles of Combinatorial testing using the PICT (Pairwise independent Combinatorial testing). By promoting a more varied set of inputs for thorough testing, PICT enhances the seed production process. Then we show how these different seeds may be easily included in the American Fuzzy Lop (AFL) tool, demonstrating how AFL can effectively use them to find and detect software flaws. This integrated technique offers a powerful yet straightforward approach to software Quality.