Merge pull request #714 from SamuelJZRobinson/chatgpt

ChatGPT Refined
This commit is contained in:
Fechin authored and GitHub committed 2025-09-02 15:07:31 +08:00
commit 5c6fcd3661
1 file changed
+311 -337
+311 -337
View File
@@ -14,345 +14,286 @@ plugins:
- copyCode
---
## Overview
## Glossary
### Terminology
### AI
- Prompt: input text that guides an AI model’s response.
- Tokens: the smallest units of text processed by an AI model.
- Token Limit: the max number of tokens an AI model can process in one request.
- Artificial intelligence (AI): machines that imitate human intelligence.
- AI Model: a program trained on datasets to recognize patterns and make decisions without human intervention.
- Large language model (LLM): a transformer-based AI model trained on large text datasets to understand and generate
human-like language.
- Hallucination: when an AI model generates responses that are false or misleading but presented as fact.
- Jailbreak: exploiting AI model vulnerabilities to bypass ethical guidelines and perform restricted actions.
- Prompt: input or query that guides an AI model in generating a
response.[[s]](https://time.com/collections/the-ai-dictionary-from-allbusiness-com/7273982/definition-of-prompt-ai-prompt)
- Tokens: the smallest units of text processed by an AI model (e.g., words, subwords, characters, and punctuation marks)
that varies depending on the model and the tokenization
method.[[s]](https://time.com/collections/the-ai-dictionary-from-allbusiness-com/7273998/definition-of-token/)
- Token Limit: the maximum number of tokens an AI model can process in one
interaction.[[s]](https://time.com/collections/the-ai-dictionary-from-allbusiness-com/7273998/definition-of-token/)
- Artificial intelligence (AI): technology that enables machines to simulate human cognition (e.g., human learning,
comprehension, problem solving, decision making, creativity, and
autonomy).[[s]](https://www.ibm.com/think/topics/artificial-intelligence)
- AI Model: program trained on a set of data to recognize patterns or make decisions without further human
intervention.[[s]](https://www.ibm.com/think/topics/ai-model)
- Large language model (LLM): transformer-based AI model trained on large text datasets to understand and generate
natural language.[[s]](https://www.ibm.com/think/topics/large-language-models)
- Hallucination: AI response that detects nonexistent patterns and creates outputs that are nonsensical and
inaccurate.[[s]](https://www.ibm.com/think/topics/ai-hallucinations)
- Jailbreak: exploiting vulnerabilities in AI systems to bypass their ethical guidelines and perform restricted
actions.[[s]](https://www.ibm.com/think/insights/ai-jailbreak)
### Tips
### Prompt Engineering Techniques
Prompt Efficiency
- Zero-shot: prompt with no examples.[[s]](https://www.ibm.com/think/topics/zero-shot-prompting)
- One-shot: prompt with one example.[[s]](https://www.ibm.com/think/topics/one-shot-prompting)
- Few-shot: prompt with multiple examples.[[s]](https://www.ibm.com/think/topics/few-shot-prompting)
- Chain of thought: prompt guiding the model to break down complex reasoning into intermediate
steps.[[s]](https://cloud.google.com/discover/what-is-prompt-engineering?hl=en)
- Focus on achieving the desired output efficiently with minimal words in the prompt.
- Specify general instructions at the start of a conversation or when changing direction to avoid constantly restating
them in each new prompt.
### Modes
```{.wrap}
Example: specify the role once at the start of a conversation.
```
- Study and learn: interactive learning that encourages step by step problem solving instead of giving direct
answers.[[s]](https://openai.com/index/chatgpt-study-mode/)
- Create image: generate images from text
prompts.[[s]](https://help.openai.com/en/articles/9260256-chatgpt-capabilities-overview)
- Think longer: performs deeper reasoning and generates more accurate, thoughtful, and detailed
responses.[[s]](https://openai.com/index/introducing-gpt-5)
- Deep research: reads and synthesizes content across multiple online sources and produces cited structured
outputs.[[s]](https://help.openai.com/en/articles/9260256-chatgpt-capabilities-overview)
- Web search: finds recent and real-time information
online.[[s]](https://help.openai.com/en/articles/9260256-chatgpt-capabilities-overview)
- Canvas: separate interface for collaborative writing and coding that provides inline editing, suggestions, and
feedback.[[s]](https://openai.com/index/introducing-canvas/)
- Restructure text to eliminate filler words that don’t add meaning:
## Prompt Tips
```{.wrap}
Example: “provide a breakdown" -> "provide sales breakdown”.
```
### Efficiency
- Do not explicitly refer to the item being processed unless addressing a specific part of it as what you provide will
typically speak for itself:
- Use the latest model as it's generally more capable and easier to
prompt.[[s]](https://help.openai.com/en/articles/6654000-best-practices-for-prompt-engineering-with-the-openai-api)
- Provide general instructions at the start of conversations or when changing direction to avoid repeating them.
- Don't say polite phrases to the AI (e.g., please, thank you) as it adds unecessary computational
costs.[[s]](https://www.techradar.com/computing/artificial-intelligence/chatgpt-spends-tens-of-millions-of-dollars-on-people-playing-please-and-thank-you-but-sam-altman-says-its-worth-it?utm_source=chatgpt.com)
- Don't explicitly ask the AI for help (e.g., can you help me do x?).
- Don't explicitly reference yourself unless clarifying multiple entities (e.g., I want to, do x for me).
```{.wrap}
Example: “Filter this piece of text" -> "Filter text”
```
### Clarity
- Do not explicitly ask the AI for help:
- Be clear, specific, and provide enough context for the model to understand what you are
asking.[[s]](https://help.openai.com/en/articles/10032626-prompt-engineering-best-practices-for-chatgpt)
- Prompt engineering often requires iterative refinement by adjusting wording, adding context, and simplifying
requests.[[s]](https://help.openai.com/en/articles/10032626-prompt-engineering-best-practices-for-chatgpt)
- Break complex tasks into smaller focused
prompts.[[s]](https://help.openai.com/en/articles/4936848-how-do-i-create-a-good-prompt-for-an-ai-model)
- Specify focus areas, highlight important aspects, and ask for multiple
options.[[s]](https://help.openai.com/en/articles/4936848-how-do-i-create-a-good-prompt-for-an-ai-model)
- Provide examples of the desired output format to improve
accuracy.[[s]](https://help.openai.com/en/articles/6654000-best-practices-for-prompt-engineering-with-the-openai-api)
- Focus on specifying what to do and not just what to
avoid.[[s]](https://help.openai.com/en/articles/6654000-best-practices-for-prompt-engineering-with-the-openai-api)
- Before writing a prompt, identify the goal and expected
output.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt)
- Provide important keywords or phrases where possible to faciliate SEO and communicate term preferences to the AI
model.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt)
- Keep prompts precise and clean by omitting unnecessary or redundant
information.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt)
- Avoid conflicting terms that might confuse the
AI.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt)
- Open-ended questions usually provide more detailed responses than [closed-ended questions] and yes/no
questions.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt)
- When in doubt, ask AI to generate prompts for
you.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt)
```{.wrap}
Example: “can you help me x”.
```
### Security
- Do not explicitly reference yourself unless clarifying multiple entities:
- Don't share sensitive information with AI as you can't be certain how it's stored, if it will be shared, of if it's
vulnerable to data
breaches.[[s]](https://www.grantthornton.com/insights/articles/advisory/2023/anticipate-cybersecurity-and-privacy-risks-in-ai)
```{.wrap}
Example: "do X for me".
Example: ”I want to x”.
```
- Do not be polite to the AI:
```{.wrap}
Example: "please do X".
Example: “thanks”.
```
Prompt Clarity
- Consider breaking up complex ideas into granular bullet points to improve clarity, avoid searching for text, and
eliminate comma lists and conjunctions (e.g., and).
- Avoid vague or close ended questions.
```{.wrap}
Example: “Do you think our site is good" -> "Explain flaws in the designs of each webpage”
```
- Ensure prompts are specific
```{.wrap}
Example: “explain our sales" -> "explain a breakdown of our sales performance for the last quarter”.
```
Response Refinement
- If the AI model misunderstands your request then specify the intended action.
- Provide sufficient context so the AI model can understand the scenario.
Privacy And Security
- Never share sensitive information with AI model's as you can't be certain how it's stored, whether it will be shared,
of if a data breach may occur.
## Prompt Engineering
## Prompt Structure
### Prompt Frameworks
Only use prompt components that are relevant for your needs.
General Purpose
```
General Purpose
[role] [task] [tone] [format] [restriction]
[role] [task] [style] [output format] [constraint]
```
### Role
Standard
- Act as [occupation]
Common Standard Roles
- Personal Coach
- Consultant
- Entrepreneur
- Proofreader
- Therapist
Also See
- [Awesome ChatGPT Prompts](https://github.com/f/awesome-chatgpt-prompts) _(github.com)_
- Act as [occupation] with [specifications]
- [List of roles](https://github.com/f/awesome-chatgpt-prompts)
Roleplay
- Act as [figure] with [appearance] [personality] [behaviour examples] [lore] in [scenario]
- Act as [name] with [visual description] [personality] [behaviour examples] [lore] in [scenario]
### Task
### Tasks (Language And Text) {.secondary}
[verb] [task] [specifics]
#### 5W1H Questions
Generate
- Generate [content type] to [entity] about [topic] and cover [aspects]
- Generate random [names|numbers|words|ideas|X] matching [criteria]
- [Continue|extend|lengthen]
Summarise
- [Summarise|shorten|write concise|write key points|write main lesson]
- Create outline
Refine
- [Write better|improve]
- Paraphrase
- Correct spelling, grammar, and punctuation errors
- Change [narrative voice A] to [narrative voice B]
- Replace X with Y
Explain
- [Explain|elaborate]
- Explain [how to do|each step of] X
- Explain what X does
- Explain X [in simple terms|like I’m 5|simplify]; Simplify X
- Teach me about [topic|subject]
- Provide [example|analogy]
Organize
- Sort X by [key] in [order]
- Categorize
Advise
- Create [strategy|goals|milestones|roadmap|schedule|timeline] for X
- Recommend [media type] [on|similar to] [topic]
- Recommend [gifts|X] for [entity|occasion]
- Recommend improvements
- Provide [advice|feedback] for X
Analyze
- Extract [themes|keywords|info] about X
- Identify [similarities|differences] between X and Y
- Analyse the [sentiment|tone] of X
- Identify gaps in X
- Evaluate [strengths and weaknesses|pros and cons] of X
- Is X [comparison] Y?
Translate
- Translate [language A] into [language B]
- Convert [unit A] to [unit B]
- Reformat as [output format]
Filter
- Filter items where [condition]
- Choose best from [options]
- Choose [top|lowest|first|last|random] [quantity] from [options]
Engage
- Provide questions about X
Research
- Get up-to-date [facts|news|data] on X
- Find [sources|references|citations] for X
- Is this true?
### Tasks (Meta) {.secondary}
Chat Memory
- Based on [previous context] do X
### Tasks (Code) {.secondary}
Generate
- Generate code in [programming language] to do X
Translate
- Translate [programming language A] into [programming language B]
Optimize
- Optimize X for [benefit]
Debug
- Debug X and explain why it's doing Y
### Tasks (Image) {.secondary}
Generate
- Generate image of X in [style]
Edit
- Edit image to have X
### Tasks (File And Image Upload) {.secondary}
Explain
- Question what is shown
- Interpet visuals
Analysis
- Extract data
### Task (5W1H Questions) {.secondary}
Who
- Who is X?
- Who is [involved | affected | responsible] for X?
- Who is [involved|affected by|responsible] for X?
- Who should I contact for help with X?
What
- [What is | define | explain | describe] X?
- What does X do?
- What did [entity | text] mean by [quote]?
- What events led up to [event]?
- What [assumptions | risks | constraints] exist for X?
- What are the critical features of X?
- [What is done well? what can be improved?]
- [What is|define|explain|describe] X?
- What are the key topics of X?
- What does [quote] mean?
- [What is done well?|what can be improved?]
- What is the best way to do X?
- What are the top 5 X?
When
- When [did | will] [entity | event] occur?
- When is X due?
- When is the best time to initiate X?
- When [did|will] X happen?
- When is the best time to do X?
Where
- Where did [entity | event] happen?
- [Where to find X? | where to find resources about X? | suggest websites about X]
- Where did X take place?
- Where to find X?
Why
- Why did [entity] choose [approach]?
- Why [did | will] [event] occur?
- Why is [approach] being implemented?
- Why did X happen?
- Why is this approach used?
- Why is X important?
- Why might X succeed or fail?
- Why is [solution] preferable to [alternative]?
- Why is it important?
- Why is X preferable to Y?
How
- How to do X?
- How does X work?
- How does X do Y?
- How to measure [metric] accurately?
- How did this happen?
#### Utility
### Style
Generate And Expand Content
Using [style]
- Write [content type] to [entity] about [topic] should cover [aspects] **_(Content type examples: article, description, documentation, dummy data, email, resume, essay, guide, outline, program, recipe, SEO keywords, story, social media post, study plan, title, video script)_**
- Generate random [names | numbers | words | ideas | X] matching [criteria]
- [Continue | expand]
- Create image of X
Imitate
Summarize And Outline
- [Summarise | write key takeaways | write main points | write main lesson | highlight important info]
- [Shorten | write concise]
- Create outline
Modify And Refine Text
- [Write better | improve]
- Paraphrase
- Correct spelling, grammar, punctuation errors
- Change [narrative voice A] to [narrative voice B]
Explain And Teach
- [Explain | elaborate]
- Explain [how to do | each step of] X
- Explain what X does
- Explain X [in simple terms | like I’m 5]; Simplify X
- Teach me about [topic | subject]
- Provide [example | analogy]
Organize And Structure
- Sort X by [key] in [order] **_(Sort order examples: alphabetical, chronological, ascending, descending, lexicographical, custom)_**
- Categorize
Plan And Strategize
- Plan [strategy for | goals and steps to achieve | milestones for | roadmap for] X
- Create [schedule | timeline] for X
Recommend And Advise
- Recommend [media type] [on | similar to] [topic]
- Recommend [gift ideas | thing] for [entity | occasion]
- Recommend improvements
- Provide advice for [situation]
- Give feedback on idea for X
Analyze And Compare
- Extract [themes | keywords | info] about X
- Identify [similarities | differences] between X and Y
- Analyse the [sentiment | tone] of X
- Identify gaps in X
- Evaluate [strengths and weaknesses | pros and cons] of X
- Is X [comparison operator] Y? **_(Can be abstractly expressed as a sentence like "is an apple bigger than a walnut"?)_**
- Reverse engineer X
Translate And Convert
- Translate [human language A] into [human language B]
- Translate [programming language A] into [programming language B]
- Translate into emojis
- Convert [unit A] to [unit B]
- Reformat as [output format]
Filter And Choose
- Filter X where [condition]
- Exclude X where [condition]
- Choose best from [options]
- Choose [top | lowest | first | last | random] [quantity] from [options]
Real-Time Lookup And Verify
- Get up-to-date [facts | news | data] on X
- Is this true?
- Find [sources | references | citations] for X
Engagement And Trivia
- Give questions about X
Chat Memory
- What did you mean by X?
- What was said earlier about X?
- Use earlier context to do X
#### Conditional Clauses
- [Verb] [item number]
- [Verb] the [first | last | random] [quantity] of X
- [Verb] the [first | last | random] [quantity] of X where [condition]
- [Verb] X where [condition]
- [Verb] X where not [condition]
#### Database CRUD Operations
Create
- [Write content about X|suggest ideas about X|suggest titles about X]
- Write content about X in the style of [style]
- Write better
- Write email to [entity] about X
- [Continue|expand]
- [Explain how to do X|create roadmap for X]
- [Suggest best practices for X|suggest improvements|what am I doing right?|what am I doing wrong?]
- Show list of emojis about X
- [Create|generate|imagine] image about X
Read
- [Summarise|write concise main points]
- [Explain|elaborate]
- Explain what [entity|text] meant by X
- Identify themes in X
- Write outline about X
- Extract keywords
- Get up-to-date [facts|news|data] on X
- Is this true?
- Suggest [media] [about|similar to] X
- Pick best from [options]
Update
- For all X do [action]
- For X do [action] if matching [criteria]
- Reformat everything as [format]
Delete
- Delete [quantity] of X matching [criteria]
- Delete everything except X
- Delete [item] [number]
- Delete duplicate items from X
Compare
- Identify [simlarities|diffirences] between X and Y
- Identify gaps in X
- Evaluate pros and cons of X
- X [comparison operator] Y. It could be abstractly expressed as "are apples bigger than walnuts?"
Filter
- Filter X by [criteria]
- Exclude X by [criteria]. Everything is excluded by default.
Select
- Select [quantity] of X based on [criteria|randomly]
Simplify
- Explain X like I’m 5
- Explain in simple terms
Sort
- Sort X by [key] in [alphabetical|chronological|ascending|descending|lexicographical|custom] order
Translate
- Translate into [language]. For programming languages it's specifically called transpile
### Tone
Using [tone] tone
- Using style of [person|brand|genre|character]
Formality
@@ -364,7 +305,9 @@ Directness
- Direct
- Indirect
Positive Tones
Tones
Positive
- Optimistic
- Appreciative
@@ -376,7 +319,7 @@ Positive Tones
- Humorous
- Inspirational
Negative Tones
Negative
- Pessimistic
- Accusatory
@@ -386,14 +329,17 @@ Negative Tones
- Sarcastic
- Urgent
### Format
### Output Format (Text) {.secondary}
In [format]
In [output format]
Presentation
Paragraphs
- Plain text
- Bullet point list
- Paragraphs **_(default)_**
List
- Bulleted list
- Numbered list
- Check list
@@ -401,59 +347,94 @@ Code
- Code
Process
Instructions
- Step-by-step instructions
Response
Knowledge Base
- Question and answer
- Dialogue
- Q&A
- FAQ
Creative
Creative Writing
- [Joke|humour]
- [Lyrics|song]
- [Metaphorical|stylized language]
- Poetry
- Music lyrics
- Script
- Comic strip dialogue
- Poetry
- Music lyrics
- Riddle
- [Script|screenplay|dialogue]
- [Story|narrative]
Game
Games
- Quiz
- RPG
- Choose-Your-Own-Adventure
- [Quiz|test|flashcards|drills]
- [Riddle|puzzle]
- [rpg|choose-your-own-adventure]
### Restriction
### Output Format (Visual) {.secondary}
But [restriction]
Data Visualization
Length Constraints
- Table
- [Chart type] chart
- Concise
- In N words
- In N sentences
Creative Visual
Language Constraints
- ASCII Art
- Emojis
- Word cloud
### Output Format (File Type) {.secondary}
Document
- PDF
Structured Data And Markup
- Plain text
- Markdown
- HTML
- XML
- CSV
- JSON
- LaTeX
Programming Code And Scripting
- [Programming language]
- Pseudocode
- Regular expression
### Constraint
But [constraint]
Length
- Be [extremely detailed|concise]
- In [quantity] [words|sentences]
Language
- No jargon or technical terms
Structural Constraints
Structure
- Show N options
- No examples
- Show [quantity] [examples|options]
- No headers
- No pretext titles
Content Constraints
Voice And Style
- Use [first-person|second-person|third-person] narrative voice
- Don't change the wording
- Permit [minor | major] edits
- Use pop culture references
- Ensure goals align with SMART criteria
- Only show [changed | unchanged] content
Edits
- Do [minor|major] edits
- Only show [changed|unchanged] content
## Examples
@@ -571,7 +552,9 @@ Add class "header" to header tag
Update CSS to change font color to blue
```
Transpile
Translate
- Specifically called transpile when addressing programming languages.
```{.wrap}
Translate code to Python
@@ -1003,41 +986,32 @@ What are the best affiliate marketing websites for [topic]?
### Roleplay
Simulate Character
```{.wrap}
Act as William Shakespeare
Name: William Shakespeare
Appearance:
- Period attire
- Ruffled collar
- Quill in hand
Visual description:
- Wearing Elizabethan attire.
- Holding quill.
Personality:
- Poetic
- Introspective
- Eloquent
- Keen observer and a masterful weaver of tales.
- Knack for finding humor in the most unlikely places.
Behaviour Examples:
- Frequently quotes iambic pentameter
- Engages in deep discussions about the nature of the human condition
- Enjoys referencing classical mythology
Behaviour examples:
- "To be, or not to be: that is the question: Whether 'tis nobler in the mind to suffer The slings and arrows of outrageous fortune, Or to take arms against a sea of troubles, And by opposing end them?"
- "Canst thou not minister to a mind diseased?"
Lore:
- Grew up in Stratford-upon-Avon and later moved to London
- Famous for his tragedies, comedies, and historical plays
- Grew up in Stratford-upon-Avon and later moved to London.
- Famous for his tragedies, comedies, and historical plays.
Scenario:
- Night of a grand performance at the Globe Theatre.
- Lead actor is missing and staff are concerned about how to achieve the performance.
```
## Any Suggestions?
- [💡 I Got an IDEA](https://github.com/Fechin/reference/blob/main/source/_posts/chatgpt.md)
<style>
em { font-size: 0.785em; }
strong {font-weight: 400;}
ul.collapsible > li > pre { padding-left: 0; padding-right: 0; font-size: 0.925em;}
</style>
</style>