Perfecting Prompts for GPT: 14 steps for harnessing the power of questions

There's a new conversationalist on the block, willing to talk with you ad infinitum, ad nauseam. It is called ChatGPT or, more recently due to a Microsoft investment, Bing. Both it and its underpinning GPT-4, however, are only as useful as the questions you ask. The better your question, the better the answer. Perfecting Prompts guides you toward this precision, with a detailed explanation of 14 different techniques to refine your query. Includes a 2-year subscription to GPTAnalytica's PromptBuilder tool. Contents:

1 Preface

2 A short history of intelligence

. . . 2.1 What is "intelligence"?

. . . 2.2 Intelligence and humans

. . . 2.3 Intelligence and computing

. . . 2.4 Artificial intelligence

. . . 2.5 Generative AI

. . . 2.6 Conversant AI

. . . 2.7 The Promethean Moment

3 Models and sources 27

. . . 3.1 Natural Language Processing (NLP)

. . . 3.2 Language Modeling (LM)

. . . 3.3 Pre-GPT Language Models

. . . 3.4 GPT Language Models

. . . . . . 3.4.1 From data to training set

. . . . . . 3.4.2 Limitations and bias

. . . 3.5 Common Crawl

. . . 3.6 WebText data set

. . . . . . 3.6.1 Test set

. . . 3.7 Wikipedia

. . . 3.8 Quality of sources

. . . 3.9 Google

4 GPT-3

. . . 4.1 Tokens

. . . 4.2 Parameters

. . . 4.3 GPT-3 and ChatGPT

. . . . . . 4.3.1 GPT-3.5

5 GPT-4

6 ChatGPT

7 Using GPT and ChatGPT in OpenAI

7.2 ChatGPT playground

7.3 Programmatic use of OpenAI

. . . 7.3.1 Import the openai library

. . . 7.3.2 An example chat API call

8 Prompt engineering

. . . 8.1 Misunderstanding in human communication

. . . 8.2 Misunderstanding in ChatGPT

. . . 8.3 Model capabilities depend on context

. . . 8.4 How to perfect your prompt

. . . . . . 8.4.1 Provide quality data

. . . . . . 8.4.2 Check your settings

. . . . . . 8.4.3 Use plain language to describe your inputs and outputs

. . . . . . 8.4.4 Show the API how to respond to any case (classify)

. . . . . . 8.4.5 Add context

. . . . . . 8.4.6 Include helpful information up-front

. . . . . . 8.4.7 Give examples

. . . . . . 8.4.8 Length of response

. . . . . . 8.4.9 Define a role

. . . . . . 8.4.10 Be more specific

. . . . . . 8.4.11 Divide a complex task into simpler tasks

. . . . . . 8.4.12 Prompt the model to explain before answering

. . . . . . 8.4.13 Ask for explanations before the answer

. . . . . . 8.4.14 Fine-tune

. . . 8.5 Additional techniques

. . . . . . 8.5.1 Selection-inference prompting

. . . . . . 8.5.2 Faithful reasoning architecture

. . . . . . 8.5.3 Least-to-most prompting

. . . 8.6 Related ideas

. . . . . . 8.6.1 Maieutic prompting

. . . 8.7 Extensions

. . . . . . 8.7.1 Self-consistency

. . . . . . 8.7.2 Verifiers

. . . 8.8 Theories of reliability

. . . . . . 8.8.1 Probabilistic graphical models

9 Act-as prompts

10 Prompt templates

11 Template libraries

12 Prompt generators

13 GPTAnalytica PromptBuilder

. . . 13.1 Obtain your OpenAI API key

. . . 13.2 Register at GPTAnalytica

. . . 13.3 Launch the tool

. . . 13.4 Build your prompt

. . . . . . 13.4.1 Select a role

. . . . . . 13.4.2 Select an action

. . . . . . 13.4.3 Select an output

. . . . . . 13.4.4 Select an input

. . . . . . 13.4.5 Write your input text

. . . . . . 13.4.6 Add above information

. . . . . . 13.4.7 Step-by-step

. . . 13.5 Refining PromptBuilder

. . . . . . 13.5.1 Comparing prompts in PromptBuilder

. . . . . . 13.5.2 Comparing prompts in OpenAI playground