Advanced Techniques
Multi-turn reasoning, tool use, long-context strategies, and prompt chaining. These techniques let you tackle complex, multi-step problems that simple prompts can't handle.
For complex analysis, break the work into stages: first extract facts, then analyse, then synthesise, then recommend. Each stage builds on the previous.
Long documents: ask the AI to first create an outline/summary of the document, then ask detailed questions referencing specific sections.
Tool use (function calling): when working with the API, define tools the AI can call (search, calculate, fetch data) to augment its capabilities.
Evaluation: ask the AI to critique its own output before finalising. 'Review your response. Is anything missing? Is anything inaccurate? Revise if needed.'
PROMPT CHAINING: Break complex tasks into sequential prompts, each building on the last output
SELF-EVALUATION: Add 'Before finalising, check your work for [specific criteria]'
MULTI-PERSPECTIVE: 'Analyse this from the perspective of [stakeholder A], then [stakeholder B], then synthesise'
FEW-SHOT LEARNING: Provide 3-5 examples of input→output pairs before giving the actual task
CONSTRAINT STACKING: Layer constraints progressively — start broad, narrow with each iteration
CONTEXT MANAGEMENT: For long conversations, periodically summarise what's been established so far
META-PROMPTING: Ask the AI to write a better prompt for a task, then use that prompt
EVALUATION FRAMEWORKS: Define rubrics for the AI to score its own output against