Day 2: Reasoning & Structure

Today covers techniques that make models think before they answer. Chain-of-thought prompting walks the model through reasoning steps. Structured output forces consistent, parseable formats. Self-consistency runs the same prompt multiple times and takes the majority answer. Meta-prompting uses an LLM to improve prompts. And templates turn prompts into reusable, version-controlled code.

Lessons

1. Chain-of-Thought Prompting , "Let's think step by step" and its variants
2. Structured Output & Format Control , JSON mode, XML tags, schema enforcement
3. Self-Consistency & Verification , running the same prompt multiple times for reliability
4. Meta-Prompting & Self-Reflection , using LLMs to write and improve prompts
5. Prompt Templates & Variables , separating prompt logic from data

Next: Day 3, Context & Optimization