Neural Fault Injection: Generating Software Faults from Natural Language

Traditional software fault injection methods, while foundational, facelimitations in adequately representing real-world faults, offeringcustomization, and requiring significant manual effort and expertise. Thispaper introduces a novel methodology that harnesses the capabilities of LargeLanguage Models (LLMs) augmented with Reinforcement Learning from HumanFeedback (RLHF) to overcome these challenges. The usage of RLHF emphasizes aniterative refinement process, allowing testers to provide feedback on generatedfaults, which is then used to enhance the LLM's fault generation capabilities,ensuring the generation of fault scenarios that closely mirror actualoperational risks. This innovative methodology aims to significantly reduce themanual effort involved in crafting fault scenarios as it allows testers tofocus on higher-level testing strategies, hence paving the way to newpossibilities for enhancing the dependability of software systems.