Use Five Whys to identify whether a wrong AI answer needs retrieval, fine-tuning, prompt, or governance changes.
The field-support assistant recommended replacing Part A, but the current service bulletin says Part B is required for devices manufactured after March. Five Whys: trace the chain until the fix points to a controllable process The trap is stopping at 'the model hallucinated' and jumping to fine-tuning. That label describes the bad output but does not identify whether source truth, retrieval, prompt, training examples, or governance failed. Why 1 Why did it recommend Part A? Because the retrieved context included an older manual that listed Part A. The first answer moves the investigation from model blame to source evidence. Why…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in