Module 2 Check · Prompting AI — The Guardrailed Classroom

01Which prompt is most likely to get a reliable, usable answer?

“tell me about benefits”“You are a VA benefits assistant. In 3 bullet points, list Post-9/11 GI Bill eligibility requirements, and cite a source for each.”“benefits???”“write everything you know”

02What is 'few-shot' prompting?

Asking the model only short questionsIncluding a few worked examples of the input→output you wantLimiting the model to a few wordsRunning the prompt a few times and averaging

03Why does asking a model to 'think step by step' often help on hard problems?

It makes the model run on more serversIt lets the model work through intermediate reasoning instead of guessing an answer immediatelyIt disables the model's safety filtersIt guarantees a correct answer

04Your prompt keeps returning prose, but you need strict JSON. Best fix?

Ask louder ('PLEASE!!!')Specify the exact output format/schema and say to return only thatRaise the temperatureNothing — models can't output JSON