国产原创

>
AI Basics

Prompting Heuristics

Using prompts effectively鈥攊.e. telling AI exactly what you need鈥攊s key to unlocking the power of AI. This page compiles different prompting heuristics鈥擳RACI, PREPARE, CREATE, and CAPTURE鈥攖hat can help improve your prompting skills. Each heuristic functions in the same way: to uncover your request鈥檚 multiple elements.

As you review the heuristics below, note that your prompt need not follow any exact order or include every single element. The essence of prompting, according to , is persona, context 补苍诲听task. The remaining elements of each heuristic can help you refine the output further.

TRACI

The TRACI Model comes from The Ohio State University (see ).

Prompt ParametersPurposeSample Prompt
TaskDescription of the primary request"List 10 evidence-based student study strategies."
RoleCharacteristics of the ideal persona"You are an experienced student success coach at The Ohio State University"
AudienceWho the response is intended for"Undergraduate students struggling with effective study skills in an introductory physics course"
CreateFormat of requested response"A concise list of study strategies with a brief summary of each strategy and references to learn more."
IntentUnderlying purpose of the output"Help physics undergraduate students learn more effective study skills to succeed in the course."

PREPARE

This heuristic was created by Dan Fitzpatrick ().

  • Prompt with a concise command.
  • Role for the AI to undertake.
  • Explicit instructions on what to do and how to do it.
  • Parameters such as format, tone, and length.
  • Ask: Tell it to ask you questions for refinement.
  • Rate: Have it grade its own response.
  • Emotion: Appeal to its emotional side.

CREATE

This model was developed by Dave Birss ().

  • Character: Give the AI a role.
  • Request: Specify the output you are looking for.
  • Examples: Give examples to exemplify the desired tone.
  • Adjustments: Add refinements in follow-up prompts.
  • Type: Define the output鈥檚 format.
  • Extra: For example, tell AI to ask you questions or explain its thought process.

CAPTURE

This model was developed by .

  • Context: Tell the LLM why we need this output.
  • Attitude: Specify the desired sentiment or tone.
  • Persona: Tell the LLM to roleplay as someone. (This often improves output.)
  • Task: Define what output the LLM should create (the core of the ask).
  • Uniqueness: Include details, adjectives and adverbs to strengthen output.
  • Requirements: Ask for a specific length, format, level of sophistication and the steps the LLM should take.
  • Explain: how is this output derived? What steps did the LLM take to arrive at an answer?