Garbage in, garbage out means that poor training data, weak instructions, or missing context tend to produce poor computational results. With LLMs, response quality depends partly on the quality and specificity of the input. section 2.2.2 section 1.2.2
Apply the principle when using AI for software development and testing:
- Question suggestions produced from potentially poor or inconsistent training data rather than accepting them blindly. section 2.2.2
- Replace vague requests with clear roles, context, constraints, quality characteristics, and concrete acceptance criteria. section 5.2.1 section 1.2.2
- Keep human judgment in control: useful output still requires review because generic or flawed suggestions can damage testing quality. section 2.2.2 section 5.2.1
From vague request to useful test ideas
“Create tests for a file upload feature” can produce a broad list of security, privacy, compliance, and performance risks that needs substantial reworking. A stronger prompt identifies the tester’s role, asks for risks affecting Functionality, Data Integrity, and Security, and supplies acceptance criteria such as supported formats, a 20MB limit, progress updates, access restrictions, and audit reporting. section 5.2.1 section 1.2.2
