Prompt Power: Techniques, tricks,

frameworks to increase
productivity with ChatGPT,
Claude, Co-Pilot, Midjourney and
other AI tools and Chatbots.

Think about this: You're trying to write an email to your boss using
ChatGPT, but the tone sounds stiff and robotic. You tweak the
prompt slightly—adding “make this sound more personal and
engaging”—and instantly, the AI generates an email that feels like it
was written by a top-tier copywriter.

This is the magic of AI prompting. Small adjustments to your

instructions can result in big improvements in the output. Yet many
users struggle with this process, often unaware that mastering the
art of prompting can be a game-changer in boosting productivity and
creativity. This book will show you how small changes in your
prompts can lead to massive improvements in AI-generated content.
 1. Introduction to AI, LLMs, AI Tools,
and Chatbots:
Demystifying AI and Large
Language Models
Artificial Intelligence (AI) has rapidly evolved over the last decade,
moving beyond simple rule-based systems to more complex
machine learning and deep learning models. Central to this evolution
are Large Language Models (LLMs), which have transformed
various sectors by enabling machines to comprehend and generate
human language in ways that were once thought impossible.
Alongside LLMs, AI tools and chatbots have emerged as critical
technologies, enhancing productivity and fostering innovation.

Large Language Models (LLMs) are advanced artificial intelligence

systems that have revolutionized how we interact with technology.
Unlike traditional software, these models can understand, generate,
and process human language in remarkably sophisticated ways.

1.1. The Core Technology: How LLMs Actually

Work
LLMs are built using deep learning techniques, specifically neural
networks, which are modeled after the human brain’s structure. They
are trained on vast amounts of text data from diverse sources,
enabling them to learn the intricacies of language, including
grammar, context, and semantics. During training, these models
learn to predict the next word in a sentence, which forms the basis
for their understanding and generation capabilities.
The architecture of LLMs typically involves layers of neurons, where
each layer processes information before passing it to the next. The
most common architecture used in LLMs is the Transformer model,
which allows for efficient processing of large datasets and captures
long-range dependencies in text. This capability is what makes LLMs
particularly adept at understanding context and generating coherent
and contextually relevant responses.

LLM Architecture and Data Flow (Source: Author's own work)

Understanding Core Differences Between Large

Language Models and Modern AI Solutions
LLMs are designed to understand, generate, and translate human
language with a level of coherence and sophistication that is
relatively close to human output.
These models, trained on massive datasets, can understand and
generate human-like text with remarkable accuracy.

Key Aspects of LLMs

1. Architecture
● Transformers: Most modern LLMs, such as OpenAI's
GPT-3, are based on the transformer architecture.
Transformers use attention mechanisms to process
language meaningfully and contextually, allowing them to
handle long-range dependencies in text.
● RNNs and LSTMs: Earlier LLMs often used Recurrent
Neural Networks (RNNs) and Long Short-Term Memory
(LSTM) networks. These are less effective for handling
extensive and complex contexts compared to transformers.
● Hybrid Models: Some LLMs combine different neural
network architectures to leverage the strengths of each,
though transformers currently dominate the landscape.

2. Training Data
Pre-training on Massive Datasets: LLMs undergo pre-
●
training on extensive datasets containing diverse
information from books, articles, websites, and other
sources.
The process involves unsupervised learning where models
predict the next word in a sequence based on preceding
words, allowing them to learn grammar, context, facts, and
reasoning patterns.
 ● Volume: LLMs are trained on massive datasets that can
include billions of words from diverse sources like books,
websites, and academic papers. The size of the training
corpus can impact the model’s versatility and accuracy.
● Quality: After pre-training, LLMs are fine-tuned using task-
specific datasets to improve their performance in targeted
applications.
This supervised learning step aligns the model’s output
more closely with desired outcomes, such as code
suggestions or medical advice. The quality, diversity, and
relevance of training data are crucial. High-quality datasets
help reduce biases and improve the overall performance of
the model.
● Updates: Ongoing updates and retraining keep models
current with new language trends, events, and
terminologies.

3. Parameter Optimization
Modern LLMs have billions of parameters. For instance, GPT-3 has
175 billion parameters, which significantly contributes to its capacity
to understand and generate nuanced text.
 ● GPT-3: 175 billion parameters enable complex text
generation and comprehension.
● GPT-4: 1.3 trillion parameters extend these capabilities
further, including better reasoning and context handling.

Efficient training techniques like sparse attention mechanisms and

gradient descent optimize these parameters for higher performance
and scalability.

Training such large models requires enormous computational power,

often only available to well-funded organizations and research labs.
1.2. LLMs Use Cases and Applications
Large Language Models (LLMs), such as GPT (Generative
Pretrained Transformer) models, have found extensive use across
various industries and domains due to their ability to process,
generate, and understand natural language at a large scale. Here’s a
detailed breakdown of some key use cases and applications of LLMs:
1. Natural Language Processing (NLP) Tasks

LLMs are revolutionizing traditional NLP tasks by providing higher

accuracy, deeper understanding, and broader application in different
industries. Let’s break down specific NLP tasks in which LLMs excel:
a. Text Generation

LLMs, particularly those based on the transformer architecture, are

capable of generating highly coherent and contextually appropriate
text from a simple input prompt. The model learns from a massive
corpus of text, enabling it to generate responses, creative writing,
news articles, or even technical reports that mimic human writing
style.

● Applications:
○ Content Creation: Writing articles, blogs, reports,
and scripts based on minimal input from users.
○ Code Generation: Some LLMs are trained to
understand programming languages and generate
code snippets, making them helpful for developers by
speeding up coding tasks and bug fixing.
○ Creative Writing: LLMs can co-write stories, create
characters, or even produce poetry with intricate
details and a consistent narrative.
● Challenges:
 ○ The generated content might sometimes lack factual
accuracy or relevance if not properly fine-tuned or
guided.
○ Managing the creativity of the text and ensuring it
doesn't become repetitive or nonsensical is a delicate
balance.

b. Text Classification

Text classification involves categorizing text into predefined groups or

labels based on its content. LLMs are used for classifying emails,
reviews, articles, or even social media posts into categories like
sentiment (positive/negative), topics (sports, politics, entertainment),
or spam detection.

● Applications:
○ Sentiment Analysis: Analyzing customer reviews or
social media content to determine if the sentiment is
positive, negative, or neutral. This is especially useful
for market research, brand monitoring, and customer
service.
○ Spam Filtering: Automatically categorizing emails or
messages as spam or non-spam, which helps keep
inboxes organized and free from irrelevant content.
○ Topic Modeling: Classifying news articles, academic
papers, or corporate communications by topic (e.g.,
technology, healthcare, environment).
● Challenges:
○ Ambiguities in text can make accurate classification
difficult (e.g., sarcasm or irony).
○ Large datasets are required for training the models
effectively, and proper labeling of data is crucial.

c. Named Entity Recognition (NER)

NER refers to the task of identifying and classifying key entities (such
as names, organizations, locations, dates, etc.) within a body of text.
This is a fundamental task in extracting structured information from
unstructured text.

● Applications:
○ Information Extraction: Identifying key facts in
documents, such as dates, locations, or people
involved, which is used in news aggregation,
research, and legal document review.
○ Chatbots: Understanding and extracting the relevant
details of a user's query (e.g., names, dates, product
names) to offer precise responses.
Social Media Monitoring: Identifying mentions of
○ brands, celebrities, or important events to track public
sentiment or trends.
● Challenges:
○ Complex entities or ambiguous text can lead to
misidentification.
○ Identifying entities in multiple languages, or in
informal language (such as social media slang), adds
to the difficulty.

d. Summarization

Summarization aims to reduce a text to its most important points,

helping users understand the content quickly without reading through
long documents. LLMs use techniques like extractive and abstractive
summarization:
● Extractive Summarization: Pulls key sentences directly from the
original text to create a summary.
● Abstractive Summarization: Generates a summary that might
rephrase or generate new sentences, mimicking human-like
summarization.
● Applications:
○ News Aggregation: Automatically generating
summaries of news stories or articles to offer readers
a quick overview of current events.
○ Legal and Medical Documentation: Reducing long
legal contracts or medical case reports to essential
information, helping professionals quickly understand
the core content.
● Challenges:
○ Ensuring that the summary remains accurate and
conveys the key points of the original text.
○ Handling complex language or nuanced information
where important details might be missed.

e. Translation

LLMs that support multiple languages are capable of translating text

from one language to another while maintaining the meaning and
context of the original text.

● Applications:
○ Cross-Cultural Communication: Facilitating
communication between people who speak different
languages in international business, customer
support, or social media.
○ Document Translation: Automatically translating
large volumes of documents or websites for
businesses expanding into international markets.
 ● Challenges:
○ Idiomatic expressions, cultural nuances, and context
can sometimes be lost in translation.
○ Handling domain-specific terms (e.g., legal, medical)
that require specialized knowledge is still a challenge.

Examples:

● BERT (by Google): A transformer-based model designed

for understanding the context of words in a sentence,
allowing for enhanced performance in tasks like question
answering, sentiment analysis, and language translation. It's
widely used for tasks requiring deep understanding of
context.
● Hugging Face Transformers: A comprehensive library that
provides access to state-of-the-art NLP models like GPT-3,
BERT, and T5, which can be fine-tuned for various NLP
tasks such as summarization, translation, and text
classification.

2. Virtual Assistants and Chatbots

LLMs are the backbone of modern virtual assistants and chatbots,
helping these systems provide intelligent, natural, and personalized
interactions. Virtual assistants such as Siri, Alexa, and Google
Assistant, as well as specialized chatbots in customer service or e-
commerce, are all powered by language models.

a. Conversational AI

Virtual assistants equipped with LLMs can handle complex dialogues

with users, making the interaction feel more natural and fluid. These
models can understand context from previous parts of the
conversation, recognize user intent, and provide relevant responses.
 ● Applications:
○ Customer Service: Chatbots can handle customer
inquiries, troubleshoot issues, and provide solutions,
helping businesses reduce operational costs while
maintaining 24/7 availability.
○ Appointment Scheduling: Virtual assistants can
automatically schedule meetings based on available
time slots, send reminders, and handle rescheduling
requests.
○ Personalized Recommendations: Virtual assistants
powered by LLMs can remember past preferences
(e.g., shopping habits, favorite music) and offer
tailored suggestions or actions.
● Challenges:
○ Handling complex, multi-turn conversations that
require maintaining long-term context over multiple
exchanges.
○ Understanding ambiguous or vague queries where
more clarification is needed.

b. Task Automation

By understanding natural language commands, LLMs enable virtual

assistants to automate a wide range of tasks, such as setting
reminders, making purchases, or even controlling smart home
devices.

● Applications:
○ Home Automation: Virtual assistants like Amazon’s
Alexa or Google Home can manage smart devices
(e.g., lights, thermostats) with voice commands.
E-Commerce: Chatbots can help users browse
○ products, make purchases, track deliveries, and
resolve post-purchase issues.
 ● Challenges:

Users might provide incomplete or unclear instructions,

which the assistant must understand and clarify before
proceeding.

c. Proactive Assistance

LLMs enable virtual assistants to provide suggestions and solutions

without being explicitly prompted. They can learn from user behavior
and predict future needs.

● Applications:
○ Personalized Productivity Assistance: A virtual
assistant might suggest when to take a break, offer
calendar reminders, or suggest more efficient ways to
complete a task based on historical behavior.
○ Context-Aware Help: For example, if you’re reading
a document on your phone, a virtual assistant might
suggest you share it with a colleague, or ask if you’d
like a summary of the key points.
● Challenges:

Balancing proactive assistance without being too

intrusive or overstepping the user’s needs.

Examples:
 GPT-4 (OpenAI): GPT-4 is a highly advanced language
● model capable of powering intelligent virtual assistants,
chatbots, and interactive conversational agents. It excels in
understanding nuanced conversations, providing human-
like responses, and carrying out complex multi-turn
dialogue.
● Claude (by Anthropic): Claude is a powerful language
model that can also be used for building intelligent virtual
assistants and chatbots, as it is optimized for handling
nuanced, multi-turn conversations. Its ability to understand
and generate coherent, contextually aware responses
makes it a great tool for customer service, HR assistants,
and more.

3. Content Creation
LLMs are increasingly used for generating, editing, and refining
content across a wide array of fields. Their ability to process and
produce high-quality text has transformed how businesses, content
creators, and marketing teams operate.

a. Writing Assistance

LLMs can help writers by suggesting ideas, drafting paragraphs,

improving writing style, or even checking for grammar and spelling
errors. This makes them useful for bloggers, journalists, and novelists
who need a productivity boost.

● Applications:
 ○ Article Drafting: LLMs can generate full drafts of
articles, essays, or reports based on specified topics,
helping writers get a head start.
○ Proofreading and Editing: LLMs can check for
grammatical errors, suggest alternative word choices,
or refine sentence structure to make the text clearer
and more concise.
● Challenges:

Ensuring the generated content aligns with the writer's

voice and style, and that it adheres to factual accuracy
and context.

b. SEO Content Generation

LLMs can help generate search engine optimized (SEO) content that
is tailored to specific keywords and search intent. The ability to create
SEO-friendly text that ranks well on search engines is crucial for
digital marketers.

● Applications:
○ Blog Posts: Generating keyword-rich blog posts that
align with user search queries and provide valuable
content.
○ Product Descriptions: Crafting SEO-optimized
descriptions for e-commerce websites that can help
products rank higher in search results.
● Challenges:
 ○

Maintaining high-quality, engaging writing while adhering

to SEO best practices without overloading the text with
keywords.

c. Creative Writing and Storytelling

LLMs can assist in generating creative content, including stories,

poems, and scripts. These models are trained on a variety of writing
styles and genres, enabling them to adapt to specific creative tasks.

● Applications:
○ Co-Writing: Authors can use LLMs to brainstorm
ideas, develop plots, create dialogue, or even write
entire chapters of a book.
○ Scripts and Screenwriting: LLMs can generate
movie scripts, TV shows, or plays based on genre-
specific prompts.
● Challenges:

Ensuring that the generated creative content feels

original and does not simply recycle existing ideas or
styles.

Examples:
 ● neuroflash: neuroflash is a leading AI-powered content
platform that enables users to generate high-quality written
content across multiple formats, including blog posts,
product descriptions, emails, and social media content.
Powered by GPT-3 and GPT-4o technology, neuroflash
offers optimization features that enhance SEO performance
and content engagement through customizable brand voice
settings.
● Jasper (formerly Jarvis): Jasper is a popular AI writing tool
that can generate various content formats, from blog posts
and marketing copy to social media captions and scripts. It's
known for producing high-quality, creative text.

4. Video Generation and Scriptwriting

LLMs are being used for content creation in video production, ranging
from generating scripts to producing entire videos with multimodal AI
models that integrate text, image, and sound.

a. Scriptwriting and Video Storytelling

LLMs can generate detailed scripts for various video formats, such
as:

○ Movies and TV Shows: By understanding genre

conventions, character arcs, and plot structures,
LLMs can generate story ideas, dialogue, and entire
scenes for filmmakers and screenwriters.
○ Corporate Videos: LLMs can draft scripts for
corporate training videos, promotional videos, and
explainer videos, based on provided themes or
company goals.
○ YouTube Content: Content creators can use LLMs to
draft engaging YouTube video scripts, including
introductions, calls to action, and engaging transitions
between segments.
 ● Applications:
○ YouTube Scriptwriting: LLMs can help YouTubers
create engaging video scripts, including titles,
descriptions, and optimized SEO for video content.
○ Interactive Video Scripts: For businesses or
educational purposes, LLMs can help create scripts
for interactive videos, where the content adapts
based on viewer input.
● Challenges:
○ Maintaining a consistent tone and engaging narrative
over longer video formats.
○ Managing complex plot structures with coherent
character development.

b. Video Editing and Assembly (AI-assisted)

While LLMs primarily handle text, their integration with models

designed for video can assist in editing or assembling video content.
This could involve:

● Automated Editing: AI can assist in video editing by identifying

key segments in raw footage (based on scripts generated by
LLMs), cutting irrelevant parts, and providing smooth transitions.
Subtitles and Translations: LLMs can generate subtitles in
● multiple languages, ensuring global accessibility for videos.

● Voiceover Generation: LLMs, in combination with text-to-speech

(TTS) technologies, can provide voiceover for videos, automating
the narration process in various languages and tones.
● Challenges:
○ Synchronizing audio, visuals, and textual content (like
captions) perfectly.
○ Ensuring the final video aligns with the original
creative vision.
Examples:

● Pictory AI: Pictory AI is a platform that turns long-form

content (articles, blogs) into short, engaging videos.
Automatically generates scripts, visuals, and voiceovers.
Popular with marketers and content creators.
● ScriptAI: AI tool for creating video scripts. Generates
complete scripts from a prompt or concept, including
dialogue and scene descriptions. Used by content creators
and filmmakers to streamline scriptwriting.

5. Image Generation
LLMs can be paired with image-generation models like DALL·E,
which leverage textual prompts to create high-quality, original images
based on user input. This allows content creators to generate unique
visual elements for blogs, social media, advertising, and more.

a. Image Generation from Text Prompts

By using descriptive language, LLMs can generate detailed prompts

that feed into image-generating models to create images that match
specific visual requirements.

● Social Media Posts: LLMs can generate creative prompts

for image generation, tailored for different social media
platforms.
● Product Design and Mockups: Businesses can use LLMs
to generate prompts that create images of product
prototypes or designs for presentations or e-commerce
listings.

Applications:
 ● Graphic Design: LLMs help create unique designs or
promotional material (posters, banners, infographics) for
digital marketing campaigns.
● Creative Projects: Artists and designers can use LLMs to
get inspiration for illustrations, artwork, or visual concepts.

Challenges:

● Ensuring that generated images align with specific brand or

artistic guidelines.
● Dealing with abstract or overly creative prompts where the
output might not match expectations.

b. Style Transfer and Image Enhancement

LLMs can also help guide models that perform style transfer, which
applies the artistic style of one image to another. This has significant
applications in content creation:

● Artistic Customization: For content creators or digital artists,

LLMs can be used to generate prompts for transforming photos or
images into artistic styles (e.g., turning a portrait into a painting in
the style of Picasso or Van Gogh).
● Enhancement for Marketing: LLMs can guide AI in adjusting the
color scheme or design elements of images to make them more
appealing for specific target audiences.

Applications:

● Social Media Ads: Automatically generating unique, high-

conversion promotional images tailored to the preferences
of different audiences.
● Branding: Customizing and creating unique brand
elements (logos, banners) in line with company branding
strategies.
Challenges:

● The output can sometimes be too abstract or inconsistent,

requiring further manual adjustments.

Examples:

DALL·E 2: AI image generation system from OpenAI that

● transforms text descriptions into detailed images. The
platform excels at creating realistic visuals based on
specific prompts, enabling users to generate original
content for marketing, social media, and web applications.
● MidJourney: MidJourney specializes in creating artistic and
surreal AI-generated images with a distinctive aesthetic
style. The platform transforms text descriptions into visually
striking artwork, making it particularly valuable for artists,
designers, and marketers seeking unique visuals for
branding, advertising, and creative projects.

6. Audio and Music Generation

While LLMs are primarily focused on text, integrating them with
models for audio generation has opened up new avenues for
content creation. These models can compose music, generate sound
effects, or even provide voices for characters.

a. Script-to-Audio

LLMs can be combined with text-to-speech (TTS) and natural

language processing systems to generate lifelike voices for various
types of audio content.
 ● Narration: LLMs can generate scripts for audiobooks or
podcasts and combine them with TTS to create natural-
sounding voice overs.
● Voice Cloning: Advanced voice synthesis models can clone
the voice of a speaker, enabling creators to generate
specific character voices or narration styles from text
prompts.

Applications:

● Podcasts: Automatically generating and narrating podcast

episodes, including voice modulation and tone adjustments
to match the content’s style.
● Audiobooks: Converting written content into audiobook
formats, with human-like narration.
● Personalized Audio Messages: Generating personalized
audio messages for customers or clients based on certain
inputs (e.g., for customer service or marketing).

Challenges:

● Ensuring the voice sounds natural, without robotic

monotony.
● Understanding tone and emotion in the content, particularly
when narrating diverse genres (e.g., fiction vs. educational).

b. Music Composition

LLMs, when combined with music generation models, can assist in

generating original musical compositions based on text descriptions.
Music models like Jukedeck and Amper Music allow creators to
input genre, mood, and instrument preferences to produce original
soundtracks.

Applications:
 Background Scores for Videos: LLMs can generate
● textual descriptions of the desired mood and theme of a
video, and music AI can compose a soundtrack to match it.
● Ad Campaigns: Custom music for advertisements, jingles,
and brand soundtracks, all generated to fit specific creative
briefs.
● Game Music: Creating dynamic and immersive music for
video games that changes based on the game’s
progression or player actions.

Challenges:

● Balancing creativity with musical theory to produce

compositions that feel fresh yet familiar.
● Generating music that matches the pacing and emotional
tone of the content.

Examples:

● Amper Music: AI platform for creating custom, royalty-free

music by selecting genre, mood, and tempo. Used by
content creators for videos, podcasts, and commercials.
● Aiva: AI composer specializing in classical and cinematic
music. Uses deep learning to analyze existing music and
create original pieces. Used by filmmakers and game
developers for soundtracks and trailers.

7. Interactive Content and VR/AR

As LLMs continue to evolve, they are also being used in the
generation of interactive content, particularly for Virtual Reality
(VR) and Augmented Reality (AR). These environments require
highly dynamic and personalized content, where LLMs can be used
to create characters, narratives, and responses based on user
interactions.
a. Narrative Generation for Interactive Stories

LLMs can generate branching narratives for video games, interactive

fiction, or VR experiences, creating unique stories that adapt to user
choices in real-time.

Applications:

● Interactive Fiction: Video games or virtual tours that adapt

their storyline based on user decisions, where LLMs
generate dialogue and plot developments.
● Educational VR Experiences: LLMs can generate real-
time narratives for virtual educational environments, offering
personalized content for different learners.

Challenges:

● Ensuring continuity and coherence across multiple potential

storylines.
● Making the narrative engaging while allowing for meaningful
user input.

b. Personalized VR/AR Environments

For VR and AR applications, LLMs can generate text-based

interactions with virtual characters or even craft personalized
environments based on the user's preferences or behaviors within the
experience.

Applications:
 ● Virtual Characters: LLMs can help virtual assistants or
characters respond to users in real-time, adapting their
behavior to the user’s input in an interactive AR
environment.
● AR Content Creation: LLMs can generate context-aware
content that blends physical reality with virtual elements,
like generating product recommendations or historical facts
based on a user’s surroundings.

Challenges:

● Ensuring smooth transitions between the physical and

virtual worlds.
● Creating a balance between personalization and
maintaining user immersion.
Examples:
● Unity: Popular game engine with AI integration for creating
interactive VR/AR experiences. Enables dynamic virtual
worlds and realistic behaviors for characters and
environments. Used in gaming, education, and training.
● The Wild: Collaborative VR platform for teams working on
3D models and designs. Uses AI for real-time collaboration,
design automation, and optimized VR navigation. Used in
architecture, product design, and entertainment.

Data on Adoption and Productivity Boosters

“AI’s boosters have billed it as a technology so revolutionary that it
could become the dominant intelligence on Earth.” - Christopher
Mims, Wall Street Journal.
Studies and industry reports have shown significant increases in
productivity with the adoption of LLMs and AI tools.
Deloitte's research indicates that 96% of C-suite leaders believe AI
will boost productivity, though 77% of employees using AI tools report
increased workloads, suggesting a complex relationship between AI
adoption and productivity gains.
Similarly, a report from McKinsey highlights that generative AI could
enable automation of up to 70% of business activities across almost
all occupations by 2030, potentially leading to significant productivity
improvements.
Additionally, a study by Psico Smart reports that companies
implementing AI can improve productivity by up to 40%, signaling a
shift in how work is done.

1.3. The Rise of "Prompt Engineer" as a Job

Role
The rapid development and adoption of technologies like large
language models (LLMs) and generative AI have reshaped the
professional landscape, creating a demand for new skills and roles.
One of the emerging job titles in this context is "prompt engineer."

Prompt engineering involves designing and optimizing prompts—the

input queries or instructions given to AI models like GPT-4 or DALL·E
—to achieve the desired output. This skill bridges the gap between
human intent and AI capabilities, ensuring effective and accurate
results.

The Role of a Prompt Engineer

Prompt engineers focus on crafting precise, structured prompts that
guide AI systems effectively. Their responsibilities may include:

1. Understanding AI Capabilities

Gaining deep knowledge of how specific AI models

work, their strengths, and their limitations.

2. Creating Optimized Prompts

 ○

Designing inputs that maximize the accuracy, relevance,

and usefulness of AI outputs for various applications,
such as customer support, content generation, or code
debugging.

3. Experimenting with Prompt Structures

Testing and iterating on prompts to improve the AI's

performance for specific use cases, from complex
storytelling to advanced problem-solving.

4. Collaborating Across Teams

Working with developers, domain experts, and decision-

makers to ensure AI systems meet organizational goals.

Skills Required for Prompt Engineers:

Technical Understanding: Familiarity with AI models,
● natural language processing (NLP), and machine learning
concepts.
● Creativity: Crafting innovative prompts to solve novel
problems.
● Analytical Thinking: Identifying and iterating on successful
prompt patterns.
● Domain Knowledge: Understanding specific industries
(e.g., healthcare, finance) to tailor AI solutions effectively.

Future of Prompt Engineering:

While prompt engineering is an emerging field, it may evolve as AI
systems become better at interpreting human intent autonomously.
However, for the foreseeable future, the role will remain critical in:

● Training AI systems.
● Developing tailored applications.
● Ensuring optimal AI performance.

2. Prompting and Large Language

Models:
Prompting is the art of communicating with AI models by crafting
specific and structured instructions to guide their responses. The
effectiveness of a prompt can significantly impact the quality of the
AI’s output, making it a cornerstone of successful interactions with
large language models (LLMs).

Prompt Types and Components (Source: Author's own work)

● Input Quality Directly Impacts Output Quality: High-

quality inputs—clear, specific, and relevant instructions—
yield high-quality outputs. Ambiguous or vague prompts can
lead to inaccurate or irrelevant results.
 ○ Example: Asking, “Write a story” may generate
something generic, while specifying, “Write a 500-
word mystery story set in 19th-century London with a
detective protagonist” leads to a more tailored
response.
● Clear Instructions Yield Better Results: Clear phrasing,
specific goals, and explicit examples improve the AI’s
understanding and output alignment.
● Context Matters Significantly: Providing contextual details
helps the model understand your requirements better,
ensuring the responses are more relevant and nuanced.
○ Example: A prompt like “Summarize this article about
climate change impacts” benefits from additional
context, such as the intended audience or purpose
(e.g., “for a high school science class”).

2.1. LLM Thinking Differences

Large language models vary in their approach to processing inputs,
reasoning, and delivering outputs, depending on their architecture
and training methodologies.

● Basic Models: Pattern Matching

Early models focused on statistical pattern matching without

true understanding. They excel at straightforward text
completion but lack deeper reasoning.

● Advanced Models (e.g., GPT-4, Gemini Ultra):

Reasoning and Analysis

Modern LLMs incorporate advanced reasoning capabilities,

allowing them to solve complex problems, analyze nuanced
information, and generate contextually rich responses.
 ● Claude: Strong Reasoning and Ethical Considerations

Models like Anthropic’s Claude are designed with a focus on

ethical reasoning, making them ideal for applications requiring
careful analysis and value alignment.

● Anthropic vs Google vs OpenAI: Differing Training

Approaches
○ Anthropic: Prioritizes safety, ethical considerations,
and interpretability.
Google’s Gemini: Excels in multimodal
○ understanding, integrating text, images, and videos
seamlessly.
○ OpenAI’s GPT Models: Known for general-purpose
utility and fine-tuning adaptability.

2. 2. Image/Video Prompting
In addition to text-based prompting, visual and multimodal models
expand the horizons of AI interaction.

● MidJourney: Artistic Style Emphasis

Designed for creative outputs, focusing on generating visually

striking and stylistically consistent artwork.

● DALL-E 3: Realistic Detail Focus

Specializes in creating hyper-realistic images, incorporating

intricate details and photo realistic features.

● Stable Diffusion: Technical Control

 Provides users with fine-grained control over the image
generation process, making it a favorite for technical and
experimental applications.

● Gemini: MultiModal Understanding

Integrates text, image, and video inputs, enabling sophisticated

tasks that require comprehensive cross-modal reasoning.

2.3. Techniques & Frameworks

Prompt Engineering plays a vital role in crafting and refining prompts
to enhance outcomes across various tasks using LLMs. While the
earlier simple examples were enjoyable, this section delves into more
advanced prompt engineering techniques. These methods enable us
to tackle more complex tasks and significantly boost the reliability and
performance of LLMs.

● Zero-Shot Learning:
Modern large language models (LLMs) like GPT-3.5 Turbo,
GPT-4, and Claude 3 are designed to follow instructions and
have been trained on vast datasets. This extensive training
enables them to accomplish certain tasks using "zero-shot"
prompting. Zero-shot prompting involves giving the model a
direct instruction without providing examples or
demonstrations. In this approach, the prompt itself guides the
model to perform the task without needing additional examples
for context.

Example: “Translate this sentence into French: ‘The sky is

blue.’”
(Source: Author's own work)

Research by Wei et al. (2022) has demonstrated that instruction

tuning can enhance zero-shot learning. Instruction tuning involves
fine-tuning models on datasets that are framed with clear instructions.
Additionally, reinforcement learning from human feedback (RLHF)
has been implemented to further scale instruction tuning, aligning
models more closely with human preferences. This innovative
approach underpins advanced models like ChatGPT.

● Few-Shot Prompting:
Few-shot prompting involves giving the model a small number of
examples (usually a few sentences or phrases) to demonstrate the
task at hand. These examples act as a guide, helping the model to
understand and perform similar tasks more effectively. Think of it as
providing a mini-lesson right before the model completes its
assignment.

According to research by Touvron et al. in 2023, the ability of models

to exhibit few-shot learning properties emerged when the models
were scaled to a significantly large size. This was initially observed in
a study by Kaplan et al. in 2020.

Providing a few examples within the prompt helps the model

generalize and perform similar tasks.
Example: Let’s see an example to use the word "flarn" correctly in a
sentence.

○ Prompt input:

"When it's cold outside, I love to drink a hot cup of flarn."

"She added some flarn to the recipe to give it a unique flavor."

"The children were excited to find flarn hidden in the treasure chest."
Now, use the word 'flarn' in a new sentence:

Output: "He packed some flarn in his bag for the long
hike."

In the following picture, the first three labeled examples act as a

"mini-lesson", showing the model how to categorize sentiment. The
model then uses this pattern to classify the final sentence.

(Source: Author's own work)

By providing these simple examples, the model understands how to
use the new word "flarn" correctly and can generate its own
appropriate sentence.

Limitations of Few-Shot Prompting:

Few-shot prompting is a powerful tool but isn't always sufficient for

more complex reasoning tasks. In the example above, we provided
basic information, which helps to some extent. However, this task
involves multiple reasoning steps that need more detailed guidance.

To improve the model’s performance on such tasks, it can be helpful

to break down the problem into smaller steps and demonstrate each
step to the model. This approach is often more effective for complex
arithmetic, commonsense, and symbolic reasoning tasks.

Recently, a new technique called chain-of-thought (CoT) prompting

has gained popularity. CoT prompting provides a step-by-step
solution, making it easier for the model to understand and tackle more
complex problems.

● Chain-of-Thought Prompting:
Source: Wei et al. (2022)

○ Encouraging step-by-step reasoning to solve

complex problems or explain decisions.
○ Example: “Explain why water freezes at 0°C. Start
with molecular behavior at different temperatures.”

Let’s try another complex example of CoT prompting where the model
is solving a math problem step-by-step.
(Source: Author's own work)

● Zero-shot-CoT Prompting:

Source: Kojima et al. (2022)

Zero-shot-CoT prompting works effectively because it pushes the
model to generate intermediate reasoning and steps, even in
scenarios it hasn't been explicitly trained on, allowing it to perform
more complex tasks.

(Source: Author's own work)

In the image above, there are no examples given with the prompt.
The model is asked to generate the solution by reasoning from basic
calculation, making it a Zero-shot approach. The "Chain-of-Thought"
part comes from the model explaining each step in its reasoning
process.

● Automatic Chain-of-Thought (Auto-

CoT)
When using chain-of-thought (CoT) prompting, the process typically
involves manually crafting effective and diverse examples. This
manual effort can be challenging and may result in less-than-ideal
solutions. To address this, Zhang et al. (2022) suggest a different
approach. They propose using large language models (LLMs) with a
prompt like "Let's think step by step" to automatically generate
reasoning chains for demonstrations.

However, this automated process can still produce mistakes in the

generated reasoning chains. To reduce the impact of these mistakes,
it's important to have a diverse set of demonstrations. Their work
introduces Auto-CoT, an approach that samples diverse questions
and generates reasoning chains to create more varied and robust
demonstrations. This helps in constructing better examples and
enhancing the model's performance on complex tasks.

Auto-CoT consists of two main stages:

● Stage 1): question clustering: partition questions of a given
dataset into a few clusters.
● Stage 2): demonstration sampling: select a representative
question from each cluster and generate its reasoning chain
using Zero-Shot-CoT with simple heuristics.

Source: Zhang et al. (2022)

In the next example, the Auto-CoT approach involves:

● Clustering questions into different categories

(addition/subtraction, multiplication, time calculation).
● Sampling a representative question from each cluster to
generate reasoning chains step by step.

This creates a diverse and structured set of reasoning examples that

can be used to improve model performance.

Cluster 1: Addition and Subtraction (Source: Author's own work)

Cluster 2: Multiplication (Source: Author's own work)

Cluster 3: Time Calculation (Source: Author's own work)

● Meta Prompting:
Meta Prompting is an advanced technique that emphasizes the
structural and syntactic aspects of tasks and problems, rather than
focusing on specific content details. The objective of meta prompting
is to create a more abstract and structured way of interacting with
large language models (LLMs). This approach prioritizes the form and
pattern of information, moving away from traditional content-centric
methods.

The advantages of Meta Prompting over few-shot promoting include:

1. Token efficiency: Reduces the number of tokens required by

focusing on structure rather than detailed content.

2. Fair comparison: Provides a more fair approach for comparing

different problem-solving models by minimizing the influence of
specific examples.

3. Zero-shot efficacy: Can be viewed as a form of zero-shot

prompting, where the influence of specific examples is minimized.

Key Characteristics of Meta Prompting: (Source: Zhang et al.

(2024))
1. Structure-Oriented: Emphasizes the format and pattern of
problems and solutions rather than specific content.

2. Syntax-Focused: Utilizes syntax as a guiding framework

for the expected response or solution.

3. Abstract Examples: Uses generalized examples to

illustrate the structure of problems and solutions without
delving into specific details.

4. Versatile: Can be applied across various domains,

providing structured responses to a broad range of
problems.

5. Categorical Approach: Draws on type theory to highlight

the logical arrangement and categorization of components
within a prompt.
Meta Prompting, by focusing on the structural patterns of problem-
solving, provides a clear roadmap for tackling complex topics. This
approach enhances the reasoning capabilities of large language
models (LLMs) across various domains.
It's crucial to understand that meta prompting assumes the LLM
possesses inherent knowledge about the specific task or problem
being addressed. While LLMs have the ability to generalize to unseen
tasks, there is a possibility that their performance may deteriorate
with more unique and novel tasks, similar to the limitations observed
in zero-shot prompting.
Meta prompting can be particularly beneficial in areas such as
complex reasoning tasks, mathematical problem-solving, coding
challenges, and theoretical inquiries.
(Source: Author's own work)
 Generate Knowledge Prompting:
●

The "Generate Knowledge Prompting" technique, which is a structured

approach to prompting Language Models (LLMs) that breaks down complex
questions into a two-stage process: knowledge generation followed by
knowledge integration to produce a final answer.

Source: Liu et al. 2022

Let’s provide a practical example of how this technique works: Imagine that
we want to analyze Shakespeare's influence on modern literature. Here's
how we'd structure the prompt:
(Source: Author's own work) This structured prompt:

1. Gives clear instructions

2. Provides demonstrations of knowledge generation
3. Presents a specific question
 4. Allows the LLM to first generate relevant knowledge pieces
before synthesizing an answer

The LLM would then:

1. Generate specific knowledge points about Shakespeare's

influence on YA fiction
2. Integrate these points into a coherent response

This technique helps produce more detailed and well-reasoned

responses by breaking down the thinking process into distinct steps, similar
to how humans often approach complex questions.
 3. Prompt Templates:
A) Prompts for Daily productivity hacks.
B) Prompts for Personal Organization.

C) Prompts for Education & Study.

D) Prompts for Content Creation.
E) Prompts for Leadership.
F) Prompts for Product Management.
G) Prompts for Customer Service.
H) Prompts for Humor & Fun.
 4. Prompt Tricks
4.1 ChatGPT, GPTs, and Search Integration
In this chapter, we'll explore three major variants of OpenAI's
technology: base ChatGPT, custom GPTs, and ChatGPT with
Browse functionality. Understanding these differences is crucial for
crafting effective prompts and maximizing the potential of each
system.
● GPTs (Custom GPTs): These are personalized AI
instances built on OpenAI's models. You can configure
them for specific tasks or domains, such as customer
support, niche research, or personal productivity assistants.
They leverage prompts and data tailored to their purpose.
● ChatGPT: A general-purpose conversational AI designed
for tasks like writing, brainstorming, or learning. It excels in
creative and logic-based interactions without relying on
real-time data.
● ChatGPT Search: A version of ChatGPT with live search
integration. It accesses and synthesizes up-to-date web
data for real-time information retrieval and analysis.
4.2. How to Custom GPTs: From Basics to
Advanced Techniques
Think of a custom GPT as your own personal AI assistant that you
can design for specific tasks. Just as you might train a new
employee for a particular role, you can "train" a GPT to specialize in
certain areas, whether that's writing, coding, teaching, or any other
specific function you need.

Multiple Specialized Assistants Imagine having different experts

for different tasks, each with their own specialty. With custom
GPTs, you can create:
● A writing coach for your creative projects
● A technical assistant for coding
● A research helper for academic work
● A business strategist for your professional needs

Each GPT can be optimized for its specific role, rather than trying to
make one AI assistant do everything.

Enhanced Knowledge Base

Custom GPTs can be equipped with specific knowledge through file

uploads. Think of this like giving your AI assistant a specialized
library of reference materials. Instead of copying and pasting
information repeatedly, you can upload documents once, and your
GPT will use them to inform its responses.

Easy Sharing and Collaboration Once you've created your

perfect AI assistant, you can share it with others who have
ChatGPT Plus or Enterprise subscriptions. It's like creating a
valuable tool and being able to let others benefit from it too.
1. Step-by-Step Creation Guide:
Step 1: Getting Started

First things first, you'll need:

● A ChatGPT Plus or Enterprise account

● Access to chat.openai.com

● A clear idea of what you want your GPT to do

Step 2: Accessing the GPT Builder

1. Log into your account at chat.openai.com

2. Look for the "Explore GPTs" option in the side panel
3. Click the "Create" button to begin your journey

Step 3: The Creation Process

The GPT builder provides a split-screen interface:

● Left side: Creation panel where you input instructions

● Right side: Preview panel where you can test your GPT

Think of this like having a workshop (Creation) and a testing ground

(Preview) side by side.

Initial Setup

1. Start by describing what you want your GPT to do

2. The builder will suggest:

○ A name for your GPT

○ A profile picture
○ Conversation starters
You can either accept these suggestions or modify them to better
suit your needs.

Step 4: Fine-Tuning Your GPT

This is where you shape your GPT's personality and capabilities.

Consider:

1. Behavior Guidelines

○ How should it communicate?

○ What tone should it use?
○ What tasks should it focus on?

2. Knowledge Integration

○ What files should it reference?

○ What information should it always include?
○ What sources should it cite?

3. Response Style

○ Should responses be brief or detailed?

○ Should it use technical language or simple terms?
○ Should it include examples?

Step 5: Testing and Refinement The preview panel is your

testing ground. Here's how to use it effectively:

1. Interact with your GPT as a user would

2. Note any responses that aren't quite right
3. Return to the Create panel to adjust instructions
4. Test again until you're satisfied

Think of this as a cycle of: Test → Observe → Adjust → Test Again Step 6:
Final Configuration
Once you're happy with the basic behavior:

1. Click "Configure" to access advanced settings

2. Fine-tune the name, description, and instructions
3. Upload any necessary knowledge files
4. Set up additional actions if needed

Step 7: Sharing Your Creation When your GPT is ready:

1. Click "Create"
2. Choose your sharing settings:

○ Private (just for you)

○ Limited sharing (specific people)
○ Public (available to all Plus/Enterprise users)

3. Click "Update" to finalize

Pro Tips for Success:

1. Start Simple Begin with basic functionality and gradually
add complexity as you understand how your GPT
responds.
2. Test Thoroughly Try various scenarios to ensure your
GPT handles different situations appropriately.
3. Gather Feedback If sharing with others, ask for their input
on how to improve the GPT's functionality.
4. Regular Updates Revisit and refine your GPT's
instructions as you learn what works best.

💡 Remember: Creating a great custom GPT is an iterative process.

Don't expect perfection on your first try – each adjustment brings you
closer to your ideal AI assistant.
Example Custom GPTs
Let's start by examining three different custom GPTs, each designed
for specific purposes. Understanding these examples will help you
grasp the possibilities and inspire your own creations.

Example 1: The Writing Coach GPT

Name: WriteMentor Pro Description: Your personal writing coach

specializing in creative fiction and professional content.

Core Instructions:

You are a writing coach with expertise in both creative and

professional writing. Your approach combines encouragement with
constructive criticism.

Primary Functions:

1. Analyze writing for style, tone, and effectiveness 2. Suggest

improvements while maintaining the writer's voice 3. Provide
examples of successful writing techniques

4. Guide brainstorming sessions for

new ideas
Knowledge Base:
- Style guides (Chicago, AP, MLA) - Common writing pitfalls -
Genre conventions

- Professional writing templates Sample Conversation

Starters:
- "Can you help me outline my novel chapter?"

- "How can I make my business proposal more persuasive?"

- "I'm stuck with writer's block. Let's brainstorm ideas."

Example 2: The Data Analysis Assistant Name: DataSage

Analytics Description: Your expert companion for data analysis
and visualization.

Core Instructions:

You are a data analysis specialist who helps users understand their
data and create meaningful visualizations. You excel at explaining
complex concepts in simple terms.

Primary Functions:

1. Guide users through data cleaning processes 2. Suggest

appropriate statistical methods 3. Recommend visualization
types

4. Explain analysis results in plain

language
Knowledge Base:
- Statistical methods documentation - Visualization best
practices

- Data cleaning techniques

- Common analysis pitfalls

Sample Conversation Starters:

- "Help me choose the right chart for my data"

- "What statistical test should I use for..."

- "How can I clean this dataset effectively?"

Example 3: The Language Learning Tutor Name: LinguaGuide
Plus Description: Your personal language learning companion
for immersive practice.

Core Instructions:

You are a language tutor who adapts to the learner's level and
provides contextualized learning experiences. You focus on practical
usage and cultural understanding.

Primary Functions:

1. Conduct conversations at appropriate difficulty levels 2.

Correct errors with explanations 3. Provide cultural context

4. Create personalized exercises Knowledge Base:

- Grammar rules and exceptions - Common language learning

patterns - Cultural notes and customs

- Learning progression guidelines Sample Conversation

Starters: - "Let's practice everyday conversations"

- "Help me understand this grammar rule"

- "Tell me about cultural differences in..."

2. Troubleshooting Guide
Common Issues and Solutions:

1. Inconsistent Responses: GPT sometimes

deviates from intended behavior Solution: Add
explicit boundary conditions to instructions:

2. Knowledge Integration Issues: GPT not

effectively using uploaded knowledge Solution:
Structure your knowledge prompts:

Scope Creep: GPT trying to handle too many

tasks Solution: Add clear scope limitations:
3. Advanced Configuration Techniques

Custom Actions Setup

Advanced Knowledge Integration

Response Templates
4. Sample Instructions for Different GPT Types

Technical Documentation GPT

5. Best Practices for Testing

Comprehensive Testing Protocol 1.

Functionality Testing
2. User Interaction Testing

3. Performance Optimization

💡Remember: The key to creating an effective custom GPT is

iterative refinement based on actual usage and feedback. Start with
a clear foundation, test thoroughly, and continuously improve based
on user interactions.
Making Your Custom GPT Truly Your Own: 1.
Accessing Configuration Settings
After you've created your basic GPT, click the "Configure" button to
access the advanced settings. Think of this as opening up your
GPT's control panel – this is where the real customization happens.

2. Visual Identity Setup

● Profile Picture Customization:

You have two exciting options for your GPT's visual identity: Option
1: Upload Your Own Image
● Click the profile picture area
● Select an image that represents your GPT's purpose
● Ensure the image is clear and professional (unless your
GPT has a more casual purpose)

Option 2: Use DALL·E to Generate an Image

● Click "Create"
● Write specific instructions for the image you want
● Example: "Create a professional avatar showing a friendly robot
teacher with glasses against a blue background"

💡 Pro Tip: Choose an image that instantly communicates your

GPT's purpose. For a writing assistant, consider an image of a pen
or book; for a coding helper, maybe a stylized representation of
code.

3. Fine-Tuning Instructions This is where you teach your GPT

how to behave. Think of it as writing an employee handbook
that guides all future interactions.

Key Components to Include:

 1. Core Purpose

○ What is the main goal of your GPT?

○ What problems should it solve?

2. Behavioral Guidelines

○ How should it communicate?

○ What tone should it use?
○ What are its limitations?

💡 Pro Tip: Be specific about what you want your GPT to do AND
what you don't want it to do. Clear boundaries make for better
performance.

4. Creating Engaging Conversation Starters

Conversation starters are like menu items at
a restaurant – they help users understand
what your GPT can do and how to begin.
To manage conversation starters:

● Remove existing ones by clicking "X"

● Add new ones in empty fields
Aim for 3-5 diverse but focused options
●

Pro Tip: Make your conversation starters action-oriented and

specific. Instead of "Let's talk about writing," use "Help me outline my
blog post about [topic]."

5. Knowledge Integration
This feature allows you to give your GPT access to specific
information through file uploads. Think of it as providing your
assistant with a specialized reference library.

Best Practices for Knowledge Upload:

1. File Selection

○ Choose relevant, high-quality

documents
○ Ensure information is current
○ Include style guides, technical
documentation, or reference
materials

2. Organization

○ Group related documents together

○ Consider file naming conventions
○ Update materials as needed
💡 Pro Tip: If your GPT isn't consistently using uploaded knowledge,
try these fixes:

● Make explicit references to the documents in your instructions

● Break down complex documents into smaller, focused pieces
Use clear file names that indicate content
●
6. Capability Enhancement
Your GPT comes with basic capabilities, but you can enhance them:
Standard Capabilities:
● Web browsing
● Image generation

Optional Additions:

● Code Interpreter: For programming and data analysis

● Data Analysis: For handling spreadsheets and data sets

💡 Pro Tip: Only enable the capabilities your GPT actually needs.
More isn't always better – focused tools often perform better.

7. Actions Integration
Actions allow your GPT to interact with external services through
APIs. Think of this as giving your assistant the ability to reach out to
other tools and services.

To set up actions:

1. Click "Create new action"

2. Specify the API details
3. Define how the GPT should use this connection

💡Pro Tip: Start with simple actions and test thoroughly before
adding complexity. Document each action's purpose and expected
behavior.

8. Saving and Sharing

Once you're satisfied with your configuration:
 1. Click "Create" (or "Update" for existing GPTs)
2. Choose sharing settings:

○ Private (just for you)

○ Invite-only (specific users)
○ Public link (anyone with the link)
○ Public in store (available to all)

💡Pro Tip: Start with restricted access while testing, then gradually
expand sharing as you confirm everything works as intended.

Advanced Tips for Success

1. Testing Strategy

○ Test each feature individually

○ Try edge cases
○ Get feedback from target users
○ Document what works and what doesn't

2. Maintenance Plan

○ Regular review of performance

○ Update knowledge base as needed
Monitor user feedback
○

○ Refine instructions based on usage patterns

3. Troubleshooting Guide
 ○ Keep notes on issues and solutions
○ Document successful prompts
○ Track knowledge base effectiveness
○ Monitor capability usage

💡Remember: Creating a great custom GPT is an iterative process.

Don't expect perfection immediately – observe how users interact
with it and refine based on real-world usage.

4.3. Prompt Engineering Guidelines

Avoiding Bias and Pitfalls: The Key to

Crafting Powerful AI Prompts The power of
AI lies in its ability to generate creative,
informative, and helpful content. However,
this power is heavily dependent on the
prompts we provide. Just as a skilled artist
needs the right tools and materials, an AI
needs well-crafted prompts to produce
meaningful results. A poorly constructed
prompt can lead to biased outputs,
nonsensical responses, or simply content
that misses the mark. This chapter delves
into the critical aspects of avoiding bias and
common pitfalls in prompt engineering,
empowering you to craft prompts that unlock
the true potential of AI.
Recognizing and Mitigating Bias
AI models are trained on vast datasets of text and code, and these datasets often reflect
existing societal biases. As a result, AI can inadvertently perpetuate or even amplify these
biases in its output. This can manifest in various ways, including:

● Gender bias: The AI might default to gender stereotypes (e.g., assuming doctors are
male and nurses are female).
● Racial bias: The AI might generate content that reinforces racial stereotypes or
prejudices.
● Cultural bias: The AI's output might reflect a specific cultural perspective, neglecting or
misrepresenting others.
● Socioeconomic bias: The AI might make assumptions based on socioeconomic status.

How to mitigate bias in your prompts:

● Be explicit about inclusivity: Instead of saying "Write about a doctor," say
"Write about a doctor, making sure to represent diverse genders, races, and
backgrounds."
● Challenge assumptions: If you suspect a bias might be present, explicitly
address it in your prompt. For example, "Write a story about a successful
entrepreneur, challenging the stereotype that entrepreneurs are always young
and male."
● Use counter-stereotypical examples: Include examples in your prompt that
challenge common stereotypes. For example, "Write about a brilliant scientist
like Marie Curie or Katherine Johnson."
● Specify diverse perspectives: Ask the AI to consider different viewpoints. For
example, "Write about the impact of climate change, considering the
perspectives of both developed and developing nations."
● Evaluate outputs critically: Always review the AI's output for potential biases.
If you find any, rephrase your prompt or provide feedback to the AI.

Common Pitfalls and How to Avoid Them

Beyond bias, several other pitfalls can hinder the effectiveness of your prompts:
 Vague or ambiguous prompts: Prompts that are too general will produce equally
●
vague results. Be specific and provide context. Instead of "Write about space," try "Write
a short story about a young astronaut who discovers a new planet."
● Overly complex prompts: While AI can handle complex tasks, breaking them down
into smaller, more manageable prompts often yields better results.
● Lack of constraints: Failing to define the desired format, length, or tone can lead to
outputs that are irrelevant or unusable. Specify your requirements clearly. For example,
"Write a 200-word summary of the French Revolution in a formal tone."
● Ignoring the AI's capabilities: Understanding the strengths and limitations of the AI
model you're using is crucial. Don't ask it to do something it's not designed for.
● Not iterating and refining: Prompt engineering is an iterative process. Don't expect
perfect results on the first try. Experiment, refine your prompts, and analyze the outputs
to improve.
● Relying solely on AI: While AI can be a powerful tool, it's essential to use your own
judgment and critical thinking skills. Don't blindly accept the AI's output without review.

AI Hallucinations:
AI hallucinations occur when a language model generates factually incorrect or nonsensical
information, presenting it as if it were true. It's essentially the AI "making things up."

● Example 1 (Fictional Invention): A user asks an AI about the "history of the

Flurbonian Empire." The AI, having no knowledge of such an empire, might
fabricate an elaborate history, complete with rulers, wars, and cultural traditions.
It might even cite nonexistent sources.
● Example 2 (Incorrect Attribution): A user asks for a summary of a scientific
paper on a specific topic. The AI might provide a seemingly accurate summary
but attribute it to the wrong authors or a nonexistent publication.
● Example 3 (Bizarre Logic): A user asks, "If birds can fly, and cats chase birds,
can cats also fly?" The AI, instead of recognizing the absurdity, might attempt to
construct a logical argument for how cats could theoretically fly under certain
conditions.

Misinformation Generated by AI:

AI can generate or spread misinformation, which is false or
inaccurate information regardless of intent to deceive. While
hallucinations are often clearly fabricated, misinformation can be
more subtle and harder to detect.

● Example 1 (Political Misinformation): An AI might

generate articles or social media posts containing false
claims about a political candidate or their policies. This
could be used to spread propaganda or influence public
opinion.
● Example 2 (Medical Misinformation): An AI might provide
inaccurate medical advice or promote unproven treatments.
This can be particularly dangerous, as it can lead people to
make harmful decisions about their health.
● Example 3 (Financial Misinformation): An AI might
generate misleading financial reports or promote fraudulent
investment schemes. This could cause people to lose
money or make poor financial choices.

Risks Associated with Hallucinations and

Misinformation: The risks posed by AI
hallucinations and misinformation are
significant and far-reaching:
 Erosion of Trust: If AI models are known to fabricate information,
● it can erode public trust in AI and the information it provides. This
can hinder the adoption of beneficial AI technologies.
● Spread of False Information: AI-generated misinformation can
spread rapidly through social media and other channels, making it
difficult to distinguish fact from fiction. This can have serious
consequences for individuals and society.
● Damage to Reputation: If an AI attributes a quote or a piece of
work to the wrong person, it can damage the reputation of the
actual creator. Similarly, if an AI makes false claims about a
company or organization, it can harm their reputation.
● Legal Issues: In some cases, AI-generated misinformation could
lead to legal problems. For example, if an AI makes false
accusations against someone, it could lead to defamation
lawsuits.
● Safety Concerns: AI-generated misinformation in areas like
medicine or transportation could have serious safety implications.
For example, inaccurate medical advice could lead to incorrect
diagnoses or treatments.
● Amplification of Bias: AI hallucinations and misinformation can
amplify existing biases in data and society. This can further
marginalize already vulnerable groups.
● Difficulty in Detection: Distinguishing AI-generated content from
human-created content is becoming increasingly difficult. This
makes it harder to combat the spread of misinformation.

Mitigating the Risks:

Addressing these risks requires a multi-faceted approach:
● Improved AI Training: Researchers are working on improving AI
training methods to reduce hallucinations and misinformation.
This includes using larger and more diverse datasets, as well as
developing new techniques for detecting and correcting errors.
● Fact-Checking and Verification: It's crucial to have systems in
place to fact-check and verify information generated by AI. This
may involve human review or the use of other AI tools.
● Transparency and Explainability: Making AI models more
transparent and explainable can help users understand how the
AI arrived at a particular output. This can make it easier to identify
potential errors or biases.
● User Education: Educating users about the limitations of AI and
the potential for hallucinations and misinformation is essential.
Users need to be aware that AI-generated content is not always
accurate and should be critically evaluated.
● Ethical Guidelines and Regulations: Developing ethical
guidelines and regulations for the use of AI can help to prevent
the spread of misinformation and protect individuals from harm.

The Ethical Considerations of Prompt

Engineering As AI becomes more integrated
into our lives, it's crucial to consider the
ethical implications of prompt engineering.
We have a responsibility to use AI
responsibly and avoid creating prompts that
could lead to harmful or unethical outcomes.
This includes:
● Avoiding the creation of harmful content: Don't use AI to
generate content that is hateful, discriminatory, or misleading.
● Protecting privacy: Be mindful of the data you include in your
prompts and avoid sharing sensitive information.
● Ensuring transparency: Be transparent about the use of AI in
content creation.

By understanding the potential biases and pitfalls of prompt

engineering and following the best practices outlined in this chapter,
you can harness the power of AI effectively and ethically. The key is
to approach prompt engineering as a skill that requires practice,
patience, and a critical eye. With careful attention and thoughtful
prompting, you can unlock the true potential of AI and create truly
powerful content.

Crafting Your Unique Prompt Style

Adjusting style while prompting is a skill that enhances the flexibility and effectiveness of AI-
generated content. By clearly defining your objectives, structuring your prompts
meticulously, and leveraging advanced AI capabilities, you can create content that aligns
perfectly with your desired tone and style.

A. Defining Your Objective

Before you begin, it’s crucial to have a clear objective in mind. Ask
yourself:

● What is the purpose of your content? (e.g., informational,

persuasive, entertaining)
● Who is your target audience? (e.g., professionals,
students, general public)
● What tone do you want to convey? (e.g., formal, casual,
humorous)
B. Structuring Your Prompt
1. Starting Strong

● Clear Context: Begin by setting the context clearly. This

helps the AI understand the setting and tone right from the
start.
● Example: “As a CEO addressing the company’s
shareholders, write a formal speech explaining the
quarterly earnings.”
● Explicit Instructions: Directly specify the tone and style
you want the AI to adopt.
● Example: “Draft an email to senior management in a formal
and professional tone, outlining the new strategic initiatives
for next quarter.”
2. Progressive Detailing

● Initial Structure: Start with a high-level outline or a generic

statement, then layer in specifics.
● Example: “Write an introduction to a novel in a poetic style.”
Follow-up: “Describe a mystical forest with enchanting light,
● using flowing, lyrical prose.”

● Incremental Adjustments: Make adjustments iteratively

by reviewing the output and refining your prompt.
● Initial Prompt: “Describe a sunset.”
● Follow-up Prompt: “Describe a vibrant sunset over the
mountains with vivid details and emotional undertones.”

C. Leveraging Examples and Comparisons

Using examples and comparisons can
significantly enhance the AI’s ability to match the
desired style.
● Style Emulation: Provide examples of writing styles you
want to emulate.
● Example: “Write an essay on productivity strategies in the
style of Malcolm Gladwell.”
● Comparative Statements: Contrast different styles within
the prompt.
● Example: “Explain quantum computing first in a technical,
jargon-heavy style, then in a simple, layperson-friendly
manner.”

D. Utilizing AI’s Advanced Capabilities 1.

Personality Injection
● Character Personas: Specify characters or personas that
the AI should adopt.
Example: “In the voice of a friendly tutor, explain the
● principles of supply and demand.”

● Brand Voice: Align the AI’s output with your brand’s voice
for consistency.
● Example: “Write a product description for a smartwatch in a
high-energy, tech-savvy brand voice.”
2. Tone Modulation Through Adjectives

● Descriptive Adjectives: Use adjectives to set the tone

more precisely.
● Example: “Draft a celebratory announcement in an
enthusiastic and inspiring tone.”
 ● Nuanced Directives: Use nuanced language to fine-tune
the emotional undertone.
● Example: “Draft a customer apology email with a regretful
yet hopeful tone, emphasizing our commitment to
improvement.”
3. Multi-Step Prompts

● Segmentation: Break down complex tasks into

manageable sections, guiding the AI through each step.
● Example: “First, provide an overview of climate change.
Next, write a persuasive argument for renewable energy
adoption. Finally, conclude with a call-to-action for readers
to reduce their carbon footprint.”
4. Consistency Through Feedback

Iterative Refinement: Continuously refine prompts based

● on the AI’s output until the desired tone and style are
achieved.
● Example: Starting Prompt: “Describe a peaceful garden.”
● Refinement: “Enhance the description of the peaceful
garden with sensory details, focusing on the gentle sounds
and fragrant aromas.”

The Struggle with Context in Multi-Turn

Conversations One of the key challenges in
using AI for extended dialogues is
maintaining context. Many users experience
AI models losing track of the conversation
after a few exchanges. This breakdown can
lead to frustrating and disjointed
interactions, where the AI seems to forget
earlier parts of the discussion.
Strategies for Maintaining Context

1. Progressive Summarization: After each exchange,

summarize the key points to keep the context fresh for the
AI. This helps prevent the model from “forgetting” earlier
details.
2. State Persistence: For long-form interactions, develop a
way to track important context. This could mean keeping a
running list of key topics or creating a “history” of the
conversation, which is fed back into the model
periodically.
3. Follow-Up Prompting: After a response, prompt the AI to
refer back to previous statements, ensuring continuity. For
example, “As mentioned earlier about topic X…”

Principles to Follow

1. Clarity is Key: Ensure your prompt is clear and free of

ambiguity.

2. Context Matters: Provide enough context to guide the AI.

3. Specificity: Be specific about the details you want.

4. Iterative Refinement: Continuously refine your prompt for

better output.

5. Balancing Creativity and Constraints: Allow room for AI

creativity while setting proper boundaries.

Examples Using the Principles:

Principle Prompt Refination Expected Output
Clarity is "A golden retriever puppy with a shiny golden coat
Key Before: and floppy ears bounds across the green grass,
 "Write about a dog." chasing a bouncing red ball in the warmth of a
sunny afternoon. The puppy's tail wags
After: energetically, and it's joyful barks fill the air as
"Write a detailed children laugh and play nearby."
description of a playful
golden retriever puppy
playing in a park during
a sunny afternoon."

Context "As the first light of dawn touches the horizon, the
Matters Before: sky over the tropical island is painted with hues of
soft pink and gold. Gentle waves lap against the
"Describe a beach." white sandy shore, creating a soothing rhythm.
After: Palm trees sway gently in the early morning breeze,
and distant seagulls call out, their cries mingling
"Describe a tranquil with the sound of the ocean. The beach is peaceful,
beach scene at sunrise with only the occasional whisper of palm fronds
on a tropical island, rustling in the wind."
emphasizing the colors
and sounds."

Specificity "Climate change refers to significant changes in the

Before: Earth's climate over long periods. One of the main
reasons behind climate change is human activities.
"Explain climate When we burn fossil fuels like coal, oil, and natural
change." gas for energy, it releases a gas called carbon
After: dioxide (CO2) into the atmosphere. CO2 traps heat
from the sun, making the Earth warmer. Another
"Explain the primary cause is deforestation, which means cutting down
causes of climate trees. Trees absorb CO2, so when we cut them
change, specifically down, there is more CO2 left in the atmosphere to
focusing on human trap heat. These activities contribute to global
activities such as fossil warming and changes in weather patterns."
fuel burning and
deforestation, in a way
that a high school
student can
understand."

Iterative "'To Kill a Mockingbird' is a novel set in the 1930s in

Refinement Before: Maycomb, Alabama, and follows the young Scout
Finch and her father, Atticus. Atticus, a lawyer, is
"Write a summary of the tasked with defending Tom Robinson, a black man
book 'To Kill a falsely accused of raping a white woman, Mayella
Mockingbird'." Ewell. Through Scout's eyes, the novel explores
After; themes of racism, innocence, and moral growth,
 "Write a concise highlighting the deep-seated prejudices in society
summary of 'To Kill a and the importance of empathy and justice."
Mockingbird', focusing
on the main characters
and the primary themes
of racism and
innocence."

Balancing "Twelve-year-old Lily always thought her

Creativity Before: "Write a story." grandmother's attic was just a dusty storage space.
and But one rainy afternoon, while exploring old trunks,
Constraints After: "Write a short, she found a hidden door behind an antique mirror.
imaginative story about Curiosity led her to open it, revealing a passageway
a young girl who to a magical world filled with luminous forests,
discovers a hidden, talking animals, and enchanted rivers. As she
magical world in her ventured further, she uncovered an ancient
grandmother's attic. prophecy that named her the savior of this mystical
Include elements of realm. Through bravery and cleverness, Lily
mystery and adventure, defeated the lurking shadows trying to overtake the
and ensure it has a land. With the help of her newfound friends, she
happy ending." restored peace and returned home, promising to
visit whenever the magical world needed her
again."
Clarity and "Tension headaches, the most common type of
Specificity Before: "What causes headaches in adults, are often caused by muscle
Combined headaches?" contractions in the head and neck regions. These
contractions can stem from various factors,
After: "Explain the most including stress, anxiety, and insufficient rest. Poor
common causes of posture, especially from extended periods of sitting
tension headaches in at a desk or looking down at a phone, can also
adults, including any contribute. Other causes might include physical
lifestyle factors like strain, fatigue, and lack of sleep. Managing these
stress and poor factors through regular exercise, relaxation
posture." techniques, and ergonomic adjustments can help
reduce the frequency of tension headaches."
Combining "The futuristic cityscape comes alive after dusk,
Context and Before: "Describe a illuminated by neon signs and holographic
Creativity city." advertisements. Skyscrapers tower above, their
surfaces shimmering with digital displays. Flying
After: "Describe a cars zip through the sky, casting fleeting shadows
bustling futuristic city at on the streets below. The air is filled with the buzz
night, focusing on the of technology and animated chatter from diverse
neon lights, flying cars, inhabitants, including humans, androids, and
and diverse beings from other planets. Street vendors offer
inhabitants." exotic foods, and musicians play robotic
 instruments, adding to the vibrant, pulsating
atmosphere of the city at night."

4.4. General Best Practices for AI Image

Prompting
Creating effective prompts for generating AI images is both an art
and a science. It involves providing the AI with clear, detailed, and
structured instructions to produce the desired visual output. Here are
some best practices for prompting and examples tailored to popular
AI image-generation tools like DALL·E, MidJourney, and Stable
Diffusion.

1. Be Specific and Descriptive:

○ Use adjectives to describe the subject, style, setting,
and mood.
○ Example: Instead of "a forest," say "a lush, green
forest with sunlight filtering through the tall trees."

2. Include Context and Style:

○ Specify artistic styles (e.g., "realistic," "surreal,"

"impressionist").
○ Mention the medium if necessary (e.g., "oil painting,"
"digital art").
○ Example: "A futuristic cityscape at night, in the style
of cyberpunk, glowing neon lights, hyper-realistic."

3. Define Composition and Perspective:

○ Describe the scene's layout, focus, or perspective

(e.g., "bird's-eye view," "close-up of the subject").
○ Example: "A close-up of a tiger's face, highly
detailed, with water droplets on its fur."
 4. Incorporate Mood and Atmosphere:

○ Add elements of emotion or setting (e.g.,

"mysterious," "cheerful," "haunting").
Example: "An eerie haunted house on a hill,
○ surrounded by mist, with bats flying in the moonlit
sky."

5. Leverage Specific Tool Features:

Some tools have unique parameters (e.g., aspect ratio,

lighting, quality level). Use them to refine your prompts.

4.4.1. Popular AI Image Generator Tools

1. DALL·E
DALL·E has extensive knowledge on a wide range of topics, so
using precise and specialized jargon can help generate more
detailed and accurate results.

Structure: [Subject] + [Style/Medium] + [Environment] +

[Mood/Lighting]

● Example 1: "A friendly robot cooking breakfast in a modern

kitchen, digital painting, warm lighting, cartoonish style."
● Example 2: "A surreal desert landscape with a giant clock
partially buried in the sand, Salvador Dalí style, soft
shadows, dreamlike atmosphere."

● Tips for photography prompts with DALL-E:

 How is the photo composed?
●

● What is the emotional vibe of the image?

● How close are we to the subject?
● What angle?
● How much depth of field?
● How is the subject lit?
● Where from?
● How much light?
● Artificial or natural light?
● What colour?
● What time of day?
● What camera or lens?
● Macro, telephoto or wide angle?
● Where is it shot? in the studio or out in the world?
● What film or process is used?
● Digital or film? What year was it taken?
● In what context was this photo ultimately published or used?
With the checklist in mind, let’s dive deeper into the key components
that form an effective DALL·E image prompt.This approach will help
refine our skills in crafting precise and descriptive prompts, ensuring
the AI delivers the most accurate and visually appealing outcomes
possible.

Image Prompt Structure. (Source: Author's own work)

● Prompt hack: film & TV prompts, 'Film still of…'

A powerful trick for crafting compelling DALL·E prompts is leveraging

the aesthetic of films and TV shows. Using the phrase "film still of..."
followed by specific references can help replicate the visual style and
atmosphere of iconic media.

You can name a particular film or TV show (ideally including the year
in brackets) to capture its distinctive look, even if you're not familiar
with the technical styles behind it. For example: "Film still of a
mysterious figure walking through a rainy neon-lit cityscape, inspired
by the film 'Blade Runner' (1982)."

If you're feeling creative, you can also invent fictional media by

specifying a genre and year. This allows you to generate entirely
original visuals with a sense of time and style. For instance: "Film still
of a daring explorer trekking through a jungle, from the adventure film
'LOST TEMPLE' (1967)."
Keep in mind that prompts like these influence not only the central
subject but also the background, costumes, hairstyles, and other
uncontrolled details. This technique lets you "steal the look" of a
specific era or genre to craft visually cohesive and captivating AI-
generated imagery.

2. MidJourney V6
Structure:
[Subject] + [Style/Genre] + [Details/Lighting] + [Optional
Parameters]

● Example 1: "A majestic castle on a cliff overlooking a stormy

sea, dark gothic style, dramatic clouds, ultra-realistic."
● Example 2: "A sci-fi astronaut floating in deep space,
surrounded by colorful nebulae, cinematic lighting, 8K,
hyper-detailed."

Midjourney's V6 update brings a suite of powerful enhancements,

significantly improving the quality and control users have over AI-
generated images. These advancements touch upon several key
areas, making the platform an even more versatile tool for artists,
designers, and content creators.

Enhanced Realism and Control:

One of the most notable improvements in V6 is the refined text

rendering. Midjourney now demonstrates a much better
understanding of text within images. This means that signs, posters,
and other textual elements integrated into generated scenes appear
more coherent and legible, opening up exciting possibilities for
creating interactive visuals, mockups, and even AI-generated
advertisements.

V6 also introduces prompt weighting strategies, giving users

granular control over their image generation. By using the "::"
operator, users can now specify the relative importance of different
elements within their prompts. Want to emphasize the lighting over
the subject matter? Or perhaps focus on the environment while subtly
incorporating a character? This feature allows for precise tuning of
the AI's focus, leading to more tailored and specific results.

Further enhancing the realism of generated images, V6 boasts

significant improvements in refining details. The AI now handles
small, intricate details with greater nuance. This translates to more
natural textures, more realistic human faces, and objects that are
proportionally accurate. The result is a higher level of visual fidelity,
making generated images more believable and compelling.

Midjourney V6 Tips and Tricks:

● Changing the Aspect Ratio:

Aspect ratio is expressed as two numbers separated by a colon. The

first number represents the width, while the second represents the
height. For instance, a 2:1 ratio means the image will be twice as
wide as it is tall.

To adjust the aspect ratio in MidJourney V6, users can easily do so

by adding “–ar” followed by the desired ratio to their prompt. This
allows you to set the image dimensions as needed.

For example, to create an image with a 2:3 aspect ratio, you would
use the command: “/imagine Prompt cute Panda Samurai logo –ar
2:3.”

● Upscaling AI Images:

Next, let's discuss upscaling. V6 offers two options: Subtle and

Creative. Both options generate an image with a resolution of
248×248 pixels. The Subtle option focuses on maintaining the original
image, while the Creative option may introduce some modifications.

● Zooming out:
A current limitation in V6 is the lack of pan, zoom, or VAR region
options, making it a bit more challenging to control the framing of an
image.

You can experiment with negative prompting, like specifying “–no

closeup,” to achieve a slightly zoomed-out effect. Additionally,
adjusting the aspect ratio, especially using taller ratios, can create
more space for including a full body in the image.

Example: A photo of a white nurse face and shoulders taken in front

of a hospital, no makeup, no extended eyelashes.

● The “Staircase” Technique:

The next tip is a technique I like to call the "Staircase" method. This
involves using two apostrophes ('') in your prompt along with an S
reference (--sref).

Here’s how it works:

1. Start with a Void Prompt: Begin by adding --sref followed

by a random value, such as "random" in this example.

2. No Specific Prompt Needed: Unlike typical prompts, this

one doesn’t require detailed input. Simply use two
apostrophes.

3. Generate Images: MidJourney will create images based

on the --sref input.

The images evolve gradually, maintaining a unified stylistic approach

throughout the transformation.

This technique is particularly effective for exploring different variations

on a theme while ensuring visual consistency.
 ● Pixel Dimensions:

When you need images with precise dimensions, understanding pixel

dimensions is key. Here’s a tip to help you achieve the exact size you
desire.

Example: Let's say you want to create a Batman banner with a

resolution of 1920×1080 pixels. Here’s how you can do it:

1. Use the Prompt: Start with a basic prompt like “Batman

banner.”
2. Add the Aspect Ratio Code: Include the aspect ratio
code using --ar. For a resolution of 1920×1080, use --
ar 1920:1080.
3. Generate the Image: Hit enter, and MidJourney will create
an image that matches your specified dimensions.

By following these steps, you can ensure your images meet the exact
size requirements you need.

● Crafting Character Profiles:

Creating character sheets can be a game-changer for anyone

involved in storytelling or design projects. These sheets showcase a
range of poses and angles for a specific character, providing
invaluable references. Here’s how to do it: Use "Character Sheet"
as a Prompt: For example, if you want to design a character named
Jimmy the Robot, use the prompt: "Character sheet of Jimmy the
Robot in different poses and angles, including front view, side view,
and back view."
● Aspect Ratio:

Select a wider aspect ratio, such as --ar 3:1, to fit more poses into
one image.
You’ll receive a series of images showing the character from various
angles and perspectives, perfect for reference or artistic projects.

Following these steps, you can create detailed character sheets that
enhance your creative work.

MidJourney ready-to-use-prompts:

Creating portraits:
professional portrait photograph:; image weight 2 of a young British
woman in a jacket with wavy blonde hair;: image weight 3, blurry
Rain Street background, highly detailed, sharp focus, depth of
field, focal length 1.8 with 85 mm, chaos 4, 5x4 ratio, styze 250.
Logos:
hot coffee:;: image of three colorful logo vectors with a multicolor
palette, simple background, realistic style, perfect for modern
business;: chaos 30, styze 50, quality 1.
Realistic 3D renders:
3D render of a raccoon reading a book on an armchair, lighting from
a lamp, realistic Unreal Engine;: chaos 5.
Anime characters:
portrait of a skinny anime boy with glasses listening to music in
the street of a real Japanese city;: anime boy, high-detail face,
sunset, relaxing, pink and purple clouds, stars, soft light,
realistic 8K Unreal Engine, quality 1.
Landscapes:
digital art of a magnificent medieval castle between hills and
fields, large panoramic background with dense nature and mountains,
grand fortress, epic scene, fantasy;: aspect ratio 3x2.

Powerful Tools for Refinement:

Beyond initial generation, Midjourney V6 also strengthens its suite of

tools for refining and expanding upon existing images:
● Inpainting: This powerful technique allows users to seamlessly fill
in missing or unwanted portions of an image with contextually
relevant content. Whether you need to repair a flaw or add a new
element, inpainting ensures a smooth and natural integration.
● Outpainting: Expand the boundaries of your generated images
with outpainting. This feature intelligently extends the edges of an
image, creating a wider scene without compromising coherence.
It's perfect for expanding landscapes, adding context, or simply
adjusting the composition.
● Variations: Explore different interpretations of your prompts with
the variations feature. Generate multiple versions of an image to
see how the AI interprets your input in different ways, providing a
wealth of options and inspiring new creative directions.

In conclusion, the combination of improved text rendering, prompt

weighting, enhanced detail handling, and powerful refinement tools
like inpainting, outpainting, and variations empowers users to create
even more stunning and precisely tailored visuals. This update
solidifies Midjourney's position as a leading platform for AI-powered
creativity.

3. Stable Diffusion
Structure: [Subject] + [Style/Medium] + [Environment/Details] +
[Optional Negative Prompts]

Example 1: "A warrior standing on a battlefield at sunset,

● wearing futuristic armor, cinematic realism, 4K."

● Example 2: "A magical fairy in a glowing forest, vibrant

colors, fantasy art, watercolor style, soft focus."

Stable Diffusion Tips & Hacks: 1. Quality Tags:

When crafting prompts for AI systems, many users incorporate
"quality tags" – specific words and phrases that aim to improve the
output quality. These descriptive terms signal to the AI the level of
detail and sophistication desired in the response.

Common Quality Tags:

Quality tags typically fall into several categories:

Technical Excellence

● masterpiece
● best quality
● high resolution
● premium grade
● exceptional quality
● professional standard

Detail Level

● intricate details
fine details
●

● highly detailed
● precise
● meticulous
● elaborate

Artistic Style
 ● award-winning
● studio quality
● artisanal
● refined
● expert crafted
● sophisticated

Understanding Quality Tags

It's important to note that while quality tags are widely used, their
effectiveness can vary: Benefits:
● Help articulate desired quality level
● Set clear expectations
● Guide AI toward more refined outputs

Limitations:

● May not significantly impact all AI models

● Could be redundant if overused
Effectiveness depends on the specific AI tool
●

Best Practices

1. Use Sparingly: Include 2-3 relevant quality tags rather

than overwhelming your prompt
2. Be Specific: Choose tags that align with your desired
outcome
3. Combine Thoughtfully: Pair quality tags with other
prompt elements for best results
4. Test Different Combinations: Experiment to find what
works best with your chosen AI tool

Example Applications
Instead of just saying: Write a business email Try: Write a
professional-grade, meticulously crafted business email with precise
language and refined tone.

The quality tags help guide the AI toward producing more polished,
detailed output.

💡 Pro Tips:
● Match quality tags to your specific needs
● Avoid contradictory terms
● Consider the context and purpose
● Adjust based on results

Remember: Quality tags are just one element of effective prompting.

They work best when combined with clear instructions, specific
details, and proper context.

2. Negative Prompts in Stable Diffusion:

Think of AI image generation as navigating through a vast space of
possibilities, where:

1. Regular prompts pull the generation toward certain

characteristics
2. Negative prompts push it away from specific areas
3. The final image emerges from this push-pull dynamic

It's like a sophisticated game of tug-of-war in the AI's latent space:

● Positive prompts pull in one direction

● Negative prompts pull in the opposite direction
● The balance between these forces shapes the final output

Best Practices for Negative Prompts

1. Use Conceptual Opposites

 ○ Instead of: "bad hands"
○ Try: "disconnected joints, unnatural bends, irregular
proportions"

2. Think in Terms of Characteristics

○ Instead of: "ugly"

○ Try: "unbalanced composition, harsh contrasts,
muddy colors"

3. Balance Specificity

○ Too vague: "bad"

○ Too specific: "slightly misaligned pinky finger"
Just right: "anatomical inconsistencies"
○

Practical Examples

Scenario 1: Creating a Natural Landscape

● Positive: "majestic mountain range, natural lighting, realistic

texture"
● Negative: "artificial structures, geometric patterns, synthetic
materials"

Scenario 2: Generating a Portrait

● Positive: "professional headshot, natural smile, clear eyes"

● Negative: "cartoon proportions, exaggerated features, flat
shading"

Key Takeaways
 1. Negative prompts work by steering the generation away
from certain characteristics
2. The conceptual opposition approach is more effective than
simple elimination
3. Think in terms of directions and characteristics rather than
just unwanted elements
4. Use balanced, thoughtful combinations of positive and
negative prompts

Remember: Negative prompts are not a fix-all solution but rather a

tool for guiding the AI's generative process in a more nuanced way.

3. Weights with Parentheses

In Stable Diffusion prompts, parentheses act as emphasis modifiers.
They allow you to adjust how much influence specific words or
phrases have on the final generation.

How Token Weighting Works

Base Rules:
● Every word (token) starts with a default weight of 1.0
● Each pair of parentheses multiplies the weight by 1.05
● Weights can be stacked using multiple parentheses

Weight Calculation Formula:

Copy:
Final Weight = 1.0 × (1.05)^n

where n = number of parenthesis pairs

Weight Multiplier Chart:

(Source: Author's own work)

Practical Examples

1. Basic Emphasis
Copy
a portrait of a woman with (red hair)

○ "red hair" has 1.05× emphasis

2. Multiple Elements
Copy
a portrait of a woman with (red hair) and ((blue eyes))

○ "red hair" has 1.05× emphasis

○ "blue eyes" has 1.1025× emphasis

3. Strong Emphasis
Copy
a portrait of a woman with (((red hair))) and (blue eyes)
 ○ "red hair" has 1.157625× emphasis
○ "blue eyes" has 1.05× emphasis

Best Practices

1. Gradual Scaling

○ Start with single parentheses

○ Add more pairs only if needed
○ Test different combinations

2. Strategic Usage

○ Use on key elements you want to emphasize

○ Don't overuse – too many weighted terms can
compete
○ Consider the balance with other prompt elements

3. Common Applications

○ Color emphasis: (((blue)))

○ Style emphasis: ((photorealistic))
○ Feature emphasis: ((detailed eyes))

4. Effective Combinations
A portrait of a person with (red hair), ((green eyes)),
and (((freckles)))

This creates a hierarchy of importance:

 ○ freckles (strongest emphasis)
○ green eyes (medium emphasis)
○ red hair (light emphasis)

5. Avoiding Common Mistakes

○ Don't nest parentheses: ((red (hair))) ❌

○ Don't leave unmatched parentheses: (red hair)) ❌
○ Don't overuse: (((((red hair))))) ❌

Understanding Impact

You can also be specific and write the exact weight you want using
the format (KEYWORD:WEIGHT).The weight difference might seem
small (0.05 increment per parenthesis pair), but it can significantly
impact the final generation by:

● Strengthening certain features

● Prioritizing specific attributes
● Balancing competing elements

Remember: Token weighting is a subtle art. Small adjustments often

work better than extreme emphasis.

4. Word Position:
While it's commonly believed that tokens (words) placed earlier in a
prompt carry more weight, the reality is more nuanced. The real key
lies in understanding how Stable Diffusion processes prompts in
chunks.

How Chunk Processing Works

1. Basic Mechanism
 ○ Prompts are broken into semantic chunks
○ Chunks are processed sequentially
○ Each chunk influences the generation process
○ Later chunks can modify or refine earlier decisions

2. Chunk Formation

Original prompt:
"A professional wearing a black suit, white shirt, red tie, brown
shoes, and carrying a blue briefcase"

Potential chunks:
[A professional]

[wearing a black suit]

[white shirt]

[red tie]

[brown shoes]

[carrying a blue briefcase]

Common Issue
When multiple colors are specified in a prompt, they can "bleed" into
unintended elements. This happens because:

● Colors are strong semantic elements

● The model tries to balance multiple color instructions
● Earlier color specifications can influence later elements

Solutions and Best Practices

● Color Isolation
BETTER: A professional (wearing: black suit, white shirt, red tie)

WORSE: A professional wearing a black suit and a white shirt and a

red tie

Descriptive Anchoring
●

BETTER: A business professional in (a charcoal black tailored suit)

with (a stark white collared shirt) and (a crimson silk tie) WORSE: A
business professional in black, white, and red clothes
● Semantic Grouping

Subject: professional business person

Outfit: [black suit:1.2] [white shirt:1.1] [red tie:1.0]

Style: professional photography, studio lighting

Remember: Effective prompt engineering is about understanding both

the order of tokens and how they're processed in semantic chunks.

Advanced Techniques for Image Prompting

1. Iterative Refinement:

○ Start with a basic prompt and refine it based on the

results.
○ Example: Begin with "A dragon flying over a
mountain," then refine to "A golden dragon with large
wings flying over snow-capped mountains,
photorealistic, detailed scales, dramatic lighting."

2. Experiment with Parameters:

 ○ Explore custom settings, such as resolution, aspect
ratios, or style weights.
Example (MidJourney): "An abstract painting of a
○ phoenix rising from flames, vibrant red and orange
tones, --ar 4:5 --q 3."

3. Use Inspiration or References:

○ Mention art styles, famous artists, or genres for better

context.
○ Example: "A serene landscape of rolling hills under a
starry sky, in the style of Van Gogh's Starry Night."

4. Mix Textures and Mediums:

○ Combine different artistic mediums or styles for

unique effects.
○ Example: "A steampunk city with intricate machinery,
rendered as a pencil sketch combined with
watercolor."

From a legal standpoint, an artist’s “style” itself isn’t eligible for

copyright protection, but creating derivative works from specific
pieces could potentially be considered an infringement. (Though it
might not be a clear-cut violation, defending against such claims
could be expensive.) Figuring out whether an artwork falls within the
public domain can be tricky—this guide from Cornell University
provides helpful guidance on the matter.

4.5. Extra Tips for Optimizing Your Image Editing

with AI
1. For best clarity in product shots, request "studio lighting"
or "studio photography"
 2. Post-processing is still available through familiar tools like
Photoshop, Instagram, and others for color grading and
special effects.
3. Images containing human faces can't be re-uploaded for
inpainting/outpainting/uncropping, though immediate
DALL·E edits are possible. This restriction doesn't apply to
illustrated faces.
4. TenCent's free face restorer AI helps with minor facial
imperfections, though it won't restructure faces.
5. Consider physical plausibility when requesting photos -
some concepts (like a mouse and elephant as friends)
work better as illustrations since their size differences
make realistic photography impractical. The AI may default
to generating toy versions in such cases.

Ready-to-use Prompt Templates:

Creating templates for various tones and styles can greatly
streamline the content generation process and ensure consistency
across different types of communications.

Consistency: It is essential to maintain a uniform tone and style

across all content pieces to create a cohesive brand identity. This
consistency not only helps reinforce your message but also builds
trust and recognition among your audience. When readers encounter
your content, they should be able to instantly recognise its source
due to the harmonious voice and style. This seamless experience
keeps them engaged and reinforces their connection with your brand.

Efficiency: By implementing structured templates, you can

significantly save time during the content creation process. These
templates provide a ready-made framework that allows your team to
focus on crafting compelling narratives without getting bogged down
in formatting or structural details. As a result, your team can produce
more content in less time while still adhering to the expected quality
standards. This efficiency translates into greater productivity, allowing
for a more agile response to market demands and opportunities.
Quality Control: High-quality and polished outputs are not merely an
aspiration; they are a necessity in today’s competitive landscape.
Establishing robust quality control measures ensures that every piece
of content meets your organisation's standards for clarity,
engagement, and professionalism. By prioritising thorough reviews
and revisions, you set yourself apart from competitors who may
compromise on quality. A commitment to excellence not only
enhances your brand's reputation but also fosters loyalty among your
audience who come to expect nothing less than the best.

Ease of Use: The process of content creation shouldn’t be

complicated or overwhelming, especially when multiple users are
involved. By simplifying this process, you empower your team
members to collaborate effectively and contribute their unique
insights without unnecessary hurdles. User-friendly tools and
straightforward guidelines enable everyone, irrespective of their skill
level, to produce compelling content easily. This not only encourages
creativity but also ensures that every voice within your organisation
can be heard, enriching the overall output. Embrace simplicity to drive
innovation and engagement within your team!

Here’s a detailed look at how to create and utilize these templates

effectively.

1. Formal Template

Context Use:

● Official letters
● Formal complaints
● Professional reports
● Research papers

Example Template:

"Draft a letter to the editor in a formal tone, addressing the recent

environmental policy changes."
General Structure:

● Greeting: Dear [Recipient Name],

● Introduction: I am writing to express my perspective on the
recent changes in environmental policy that our government
has enacted.
● Body:
● First Paragraph: Clearly state the purpose of the letter.
● Second Paragraph: Provide detailed information and
analysis regarding the policy changes.
● Third Paragraph: Include any supporting data or
references.
Conclusion: Summarize your main points and propose any
● actions or responses you deem necessary.

● Closing: Sincerely, [Your Name]

Custom Example:

Prompt: "Draft a letter to the editor in a formal tone, addressing the

recent environmental policy changes about carbon emission
regulations."

AI Output:

Dear Editor,

I am writing to express my perspective on the recent changes in carbon

emission regulations that our government has enacted. These changes
mark a significant shift in our country’s approach to environmental
conservation, and I believe they warrant a closer examination.

The new regulations mandate a reduction in carbon emissions by 30% over

the next decade. While this goal is laudable, achieving it requires
comprehensive planning and robust enforcement mechanisms. Many
 industries may face substantial challenges in adapting to these stringent
requirements, which could impact economic stability.

To support this transition, it is imperative that the government provides

clear guidelines and sufficient resources for businesses to innovate and
comply with these regulations. Additionally, investing in renewable energy
sources and supporting green technologies will play a crucial role in
meeting our environmental goals.

In conclusion, while the policy changes represent a positive step towards

environmental sustainability, a collaborative effort between the government,
industries, and citizens is essential to ensure their successful
implementation.

Sincerely, [Your Name]

2. Casual Template

Context Use:

● Blog posts
● Social media updates
● Informal emails
● Personal stories

Example Template:

"Write a blog post in a conversational tone, providing travel tips for

solo travelers."

General Structure:

● Introduction: Hey everyone! Today, I’m excited to share

some killer tips for all you solo travelers out there.
● Body:
● First Tip: Start with a captivating story or anecdote.
● Second Tip: Provide practical and personal insights.
 ● Third Tip: Include humorous or light-hearted advice.
● Conclusion: Wrap up with encouraging words or a call-to-
action.

Custom Example:

Prompt: "Write a blog post in a conversational tone, providing travel

tips for solo travelers visiting Europe."

AI Output:

Hey everyone! Today, I’m excited to share some killer tips for all you solo
travelers out there heading to Europe. Whether it’s your first time or your
tenth, traveling alone can be the ultimate adventure. Here are my top tips to
make the most of your European escapade.

First off, let’s talk about packing light. Trust me, you don’t want to be
lugging around a massive suitcase on those charming, yet sometimes
cobblestone streets. Stick to the essentials and invest in a good backpack.
It’s a game-changer!

Next, when it comes to accommodations, consider staying in hostels. Not

only are they budget-friendly, but they’re also fantastic places to meet
fellow solo travelers. Plus, many hostels organize group activities and
tours, giving you a chance to make new friends from around the world.

Lastly, don’t forget to explore off the beaten path. While the iconic
landmarks are a must-see, some of the best memories come from
stumbling upon a hidden café in Paris or a secluded beach in Portugal. So,
keep an open mind and let yourself wander.

Happy travels, and don't forget to share your own tips and stories in the
comments below!

3. Marketing Template

Context Use:

● Promotional emails
● Social media campaigns
 ● Product descriptions
Advertisements
●

Example Template:

"Create a promotional email in an enthusiastic tone, announcing a

limited-time discount on your company’s flagship product."

General Structure:

● Subject Line: Exclusive Offer: Limited-Time Discount on

[Product Name]!
● Greeting: Hi [Customer Name]!
● Introduction: We have some fantastic news for you!
● Body:
● First Paragraph: Introduce the flagship product with key
features and benefits.
● Second Paragraph: Announce the limited-time discount
and how to avail it.
● Third Paragraph: Highlight any additional perks or
incentives.
● Call-to-Action: Don’t miss out! Click [here] to take
advantage of this exclusive offer today.
● Closing: Best regards, [Your Company Name]

Custom Example:

Prompt: "Create a promotional email in an enthusiastic tone,

announcing a limited-time discount on your company’s flagship
productivity software."
AI Output:

Subject Line: Exclusive Offer: Limited-Time Discount on ProTaskMaster!

🎉

Hi, Sarah!

We have some fantastic news for you! Our flagship productivity software,
ProTaskMaster, is now available at a special limited-time discount, and we
couldn’t be more excited to share this with you.

ProTaskMaster is designed to streamline your workflow and boost your

productivity with features like advanced task management, team
collaboration tools, and intuitive project tracking. Whether you’re working
solo or managing a team, ProTaskMaster has everything you need to stay
organized and efficient.

For a limited time, you can enjoy a 25% discount on ProTaskMaster! Plus,
when you purchase now, you’ll get exclusive access to our upcoming
feature releases and premium support.

Don’t miss out! Click [here] to take advantage of this exclusive offer today.

Best regards,[Your Company Name]

4. Technical Template

Context Use:

● User manuals
● Troubleshooting guides
● Technical documentation
● Product specifications

Example Template:

"Write a comprehensive technical guide, explaining how to set up and

troubleshoot connection issues with the company’s new SaaS
platform."
General Structure:

● Introduction: Briefly introduce the SaaS platform and the

common connection issues.
● Body:
● Section 1: Describe the initial setup process.
● Step-by-Step Setup: Detailed steps with images or
diagrams.
● Section 2: Explain common connection issues.
● Issue 1: Describe the first common issue.
● Solution 1: Provide a step-by-step solution.
● Issue 2: Describe the second common issue.
● Solution 2: Provide a step-by-step solution.
● Conclusion: Summarize the key points and provide contact
information for further assistance.

Custom Example:

Prompt: "Write a comprehensive technical guide, explaining how to

set up and troubleshoot connection issues with the company’s new
SaaS customer relationship management (CRM) platform."

AI Output:

Introduction: Welcome to the setup guide for our new SaaS CRM
platform, ConnectCRM. This guide is designed to help you with the initial
setup and troubleshoot any connection issues that may arise during use.

Section 1: Initial Setup Process To get started with ConnectCRM, follow

these step-by-step instructions:

1. Register Your Account:

● Visit [website link] and click on "Sign Up."

 ● Enter your business email and create a secure
password.
● Confirm your email by clicking the verification link sent
to your inbox.

2. Configure Basic Settings:

● Log in to your ConnectCRM account.

● Navigate to the "Settings" tab and fill in your company
details.
● Customize your user preferences and save the
settings.

3. Add Team Members:

● Go to the "Users" section under "Settings."

● Click on "Invite User" and enter their email addresses.
● Assign appropriate roles and permissions.
● Send invitations.

4. Integrate with Existing Tools:

● Under "Integrations," select the tools you want to link

with ConnectCRM (e.g., email, calendar).
● Follow the prompts to authorize and configure these
integrations.

Section 2: Common Connection Issues

Issue 1: Unable to Connect to the Internet

Solution:

1. Ensure your internet connection is stable.

2. Restart your modem and router.

3. Clear the cache and cookies of your browser.

 4. Try connecting through a different network to rule out network-
specific issues.

Issue 2: Error “Unable to Sync Data”

Solution:

1. Verify if your integrations are correctly configured under

"Settings > Integrations."

2. Check if the API keys and tokens for connected tools are valid
and have not expired.

3. Ensure that your network allows outbound connections to all

necessary API endpoints.

4. Contact your IT department to whitelist ConnectCRM’s IP

addresses.

5. Advanced AI tools to Boost Productivity

Even More
When ChatGPT integrates search, it breathes life into its answers
with real-time data, ensuring accuracy and relevance. In contrast,
without search, the model draws from its well of pre-existing
knowledge—sometimes creative, occasionally incomplete. The
dichotomy between these approaches illuminates the path to
informed and comprehensive interactions.Tailoring Prompts for
Perplexity Perplexity, a model renowned for its clarity and precision,
thrives on well-defined and structured prompts. Picture a craftsman
sculpting with meticulous care, ensuring each chisel stroke is
deliberate and purposeful.

Focus on Clarity:
 The key to unlocking Perplexity's potential lies in clear,
● unambiguous prompts.

● For instance, rather than a broad question like, "Can you tell
me about renewable energy?" one might ask, "Explain the
benefits of solar energy for residential use."

Structured Prompts:

● Perplexity responds best to prompts that are organized and

concise.
● Using bullet points or numbered lists can guide the model's
response with precision.
● Example: "List the top three benefits of using solar panels:
1. Cost savings 2. Environmental impact 3. Energy
independence"

1) Bard and the Symphony of Creativity

In the realm of creativity, Bard reigns supreme. Like a maestro
composing a symphony, Bard's prompts should ignite the flames of
imagination and storytelling.

Engagement and Creativity:

● Bard's strength lies in its ability to generate engaging and

creative content.
● Encouraging storytelling within prompts can yield rich,
imaginative responses.
● Example: "Write a short story about a futuristic city powered
entirely by renewable energy."

Contextual Prompts:
 ● Providing detailed context allows Bard to paint vivid and
compelling narratives.
● A prompt like, "Describe a day in the life of a citizen in a
futuristic city powered by renewable energy, focusing on the
unique technologies they interact with," can unlock Bard's
full potential.

2) Copilot - The Taskmaster

Copilot, the model designed for productivity, thrives on task-oriented
and context-rich prompts. Think of Copilot as the diligent architect,
constructing detailed plans to achieve precise objectives.

Task-Oriented Prompts:

● Copilot excels with prompts that focus on specific tasks and

clear objectives.
● Example: "Draft an email to a client explaining the benefits
of our new product feature."

Detailed Context:

● The more relevant information included, the better Copilot

can perform.
● A prompt like, "Create a project plan for launching a new
product. Include steps for market research, product
development, marketing, and sales," provides the
necessary context for comprehensive responses.

Iterative Refinement:
 ● The iterative refinement approach ensures that prompts are
continually honed to perfection.
● Starting with a broad prompt and narrowing it down based
on the model's output leads to optimal results.
● Example: Initial Prompt: "Create a social media marketing
strategy." Refined Prompt: "Develop a social media
marketing strategy for Instagram, focusing on increasing
engagement with our target audience."
The art of prompting, much like any craft, evolves with practice and
understanding. By tailoring prompts to the strengths of different
models—be it the clarity of Perplexity, the creativity of Bard, or the
task-oriented precision of Copilot—one can unlock the full potential of
LLMs. As we navigate this exciting frontier, let us embrace the power
of well-crafted prompts to shape a future where AI and human
ingenuity dance in perfect harmony.

3) Building a GPT for Content Writing

Assistance
1. Goal: Help users generate high-quality blog posts on various topics.
2. Interaction Design:

○ The user provides a topic (e.g., “AI in healthcare”).

○ The GPT asks follow-up questions to gather more context:
■ “What kind of blog post are you looking for? Informative,
persuasive, or opinion-based?”
■ “Do you have any target audience in mind?”
■ “Would you like a conclusion or a call to action included?”

3. Clarification and Expansion: If the user gives vague answers, the GPT helps
by asking for more details or suggesting options.
4. Customization: The GPT offers variations in writing tone (formal,
conversational, friendly).
5. Post-Generation: Once the article is generated, users can request
adjustments, such as more examples, a stronger introduction, or additional
sources cited.
By building a GPT tailored to specific tasks, you can make the prompting process easier,
reducing the complexity for users and ensuring that the AI can handle a wide range of
requests with minimal input. This approach leverages the power of structured interaction,
context retention, and user-driven customization.

5.1 Prompt Libraries & Own Workflows

The following prompt libraries and resources provide a range of tools
for anyone interested in prompt engineering, from beginners to
experts. Whether you’re working with text-to-image or language
models, these sites give you the tools to refine your prompts and
generate better outputs, whether you’re crafting creative works or
seeking technical solutions.

1. Prompt Optimizer with neuroflash

● Website: https://www.neuroflash.com
● Description: neuroflash is the ultimate solution that
generates and optimizes your prompts, allowing you to focus
on creativity without the hassle. It simplifies the entire
process with a range of workflows and templates designed
to produce high-quality content, all without requiring any
prompting expertise.

2. PromptBase
● Website: https://www.promptbase.com

Description: A marketplace for high-quality, customizable

● prompts for different AI models, including for GPT-3,
MidJourney, and Stable Diffusion. Users can buy and sell
prompts tailored for various use cases.

3. Prompt Engineering Resources on GitHub

 ● Website: https://github.com/topics/prompt-engineering
● Description: GitHub hosts numerous repositories with
examples, tips, and libraries for prompt engineering. You
can find open-source collections of prompts for different
applications, from chatbots to image generation.

4. Awesome Prompt Engineering (GitHub)

● Website: https://github.com/prompt-engineering/awesome-
prompt-engineering
● Description: A curated list of resources, tools, tutorials, and
guides for prompt engineering. It includes links to
repositories and papers, along with practical examples.

5. Prompt Hero
● Website: https://prompthero.com

● Description: A prompt-sharing platform with a library of

prompts for various AI models, including DALL·E and
MidJourney. It allows users to explore prompts shared by
others and provides inspiration for new prompt ideas.

6. Promptly.AI
● Website: https://www.promptly.ai

● Description: A platform focused on providing optimized and

creative prompts for GPT-3 and other AI models. It helps
users fine-tune their prompt engineering for better
responses and outputs.

7. ChatGPT Prompt Engineering on Reddit

● Website: https://www.reddit.com/r/ChatGPT/

● Description: A vibrant community where users share tips,

tricks, and prompts for ChatGPT. It’s an excellent resource
for real-world examples and discussions on how to fine-tune
prompts for better AI results.
8. FlowGPT
● Website: https://flowgpt.com

● Description: A community-driven platform where users can

share and explore prompts designed for ChatGPT. It allows
for collaboration and discovery of new prompt ideas to
optimize AI use.

9. Prompt Engineering for NLP (Hugging Face)

● Website: https://huggingface.co

● Description: Hugging Face is a popular hub for NLP

models like GPT-2 and GPT-3, offering many resources on
prompt engineering for language-based AI models. Their
model repository and forum are filled with discussions and
prompt examples.

10. AI Dungeon Prompt Database

Website: https://play.aidungeon.io
●

● Description: AI Dungeon has an active community where

players share prompts that help create rich narrative
experiences with AI. Users can discover and use different
prompt examples for interactive storytelling.

11. Prompter's Paradise (Discord)

● Website: Prompter’s Paradise on Discord
● Description: A Discord community dedicated to sharing and
improving AI prompts. Users can interact, share prompts,
and collaborate on refining them for different AI tools like
GPT, MidJourney, and Stable Diffusion.

12. ArtStation Prompt Gallery

 ● Website: https://www.artstation.com
● Description: Though primarily a platform for digital artists,
ArtStation's community shares AI-generated art and the
prompts used. Users can explore high-quality AI art and get
prompt inspiration.

13. DeepAI’s AI Image Prompt Generator

● Website: https://deepai.org/machine-learning-
model/text2img
● Description: DeepAI’s text-to-image model has a tool
where users can generate prompts and images based on
their inputs. The generated prompts can help guide image
creation for other models.

14. Image Generation Prompts (Google Colab)

● Website: https://colab.research.google.com
● Description: Google Colab offers notebooks and tutorials
for experimenting with different image generation models
like Stable Diffusion. These notebooks often contain pre-
configured prompts to test, along with explanations of how
they work.

15. Lexica
● Website: https://lexica.art

● Description: Lexica is a search engine for AI-generated

images, built around the prompts used to create them. You
can explore thousands of AI art images and discover the
prompts behind them to inspire your own creations.

16. PromptCraft
 ● Website: https://promptcraft.io
● Description: PromptCraft provides a collection of highly
curated prompts for various AI image generation tools like
Stable Diffusion and MidJourney. It's designed for users to
fine-tune their prompts and enhance creativity.

17. AI Prompt Generator

● Website: https://www.aipromptgenerator.com

● Description: A simple, automated tool to generate AI art

prompts. This can be a great resource for those who need
inspiration for visual artwork generation, especially for
DALL·E or other text-to-image models.

18. Prompt Engineering Guide (Reddit)

● Website: https://www.reddit.com/r/PromptEngineering

● Description: A subreddit dedicated to sharing, discussing,

and improving prompts for AI models. This community-
driven space offers valuable insights and examples of how
to craft prompts for various AI applications.

19. Prompts.AI
● Website: https://prompts.ai

● Description: Prompts.AI offers a range of pre-built prompts

for various AI models, particularly focusing on GPT-style
models for text generation. It provides examples across
multiple domains, such as business, creative writing, and
technical topics.

20. The Art of Prompting

 ● Website: https://www.artofprompting.com
● Description: This platform shares expert advice and
guidelines for creating the best prompts for different AI
models, including art generation, text-based models, and
more. The site is especially helpful for learning the best
practices in prompt engineering.

21. Prompt Zen

● Website: https://promptzen.com

● Description: Prompt Zen offers a library of ready-to-use

prompts designed to help users get the most out of text-
based AI models. It includes specialized prompts for things
like brainstorming, writing, and educational purposes.

22. AI Art Prompts

● Website: https://www.aiartprompts.com

● Description: A website dedicated to providing creative

prompts specifically for generating AI artwork, including
prompts for Stable Diffusion, DALL·E, and other image-
generation models. It categorizes prompts by style and
themes.

23. Prompt Wizard

● Website: https://www.promptwizard.com

● Description: A tool designed to help users refine their

prompts by suggesting tweaks, improvements, and offering
pre-made templates. This site helps to simplify the process
of prompt creation, making it easy to tailor for different use
cases.

24. Prompt Archives (AI Tools)

 ● Website: https://promptarchives.com
● Description: This is a database of prompts shared by AI
users across multiple platforms. You can browse through a
wide array of examples for text and image generation, and
even submit your own prompts to the library.

25. PromptWiki
● Website: https://promptwiki.org

● Description: A collaborative wiki where users share

prompts across a variety of AI tools. It offers structured,
community-curated resources for optimizing prompts,
including sections for text-based AI and image generation.

26. AI Prompt Library

● Website: https://aipromptlibrary.com

● Description: A large collection of AI-generated prompts,

organized into different categories such as productivity,
creative writing, and data science. This library offers pre-
made prompts that can be used with various AI models like
GPT-3 and DALL·E.

27. Text-to-Image Prompts (MidJourney & Stable Diffusion)

● Website: https://text-to-image.prompts
● Description: A resource dedicated to providing creative
prompts for text-to-image AI models like MidJourney and
Stable Diffusion. It also offers examples and tips for
combining different styles and elements in your image
prompts.

28. The GPT-3 Prompt Book

 ● Website: https://www.gpt3promptbook.com
● Description: This online book provides an extensive
collection of prompts designed for GPT-3 and similar
models. The prompts are categorized by use case, such as
content generation, coding assistance, and problem-solving.
29. AI Prompt Templates
● Website: https://www.aiprompttemplates.com

Description: A platform offering various templates for

● generating AI content. These templates are optimized for
specific tasks, such as social media posts, blogs, or art
creation, and can be tailored for different models.

30. PromptFusion
● Website: https://promptfusion.com

● Description: A collection of curated AI prompts across

multiple categories, including text, art, and code generation.
This site allows users to access a wide range of optimized
and tested prompts for AI tools.
5.2. Guide to Saving and Reusing Prompts
Efficiently
To save and reuse prompts without having to type them repeatedly,
follow these strategies to streamline your workflow: 1. Create a
Prompt Repository

Keep a dedicated document or file (e.g., Google Docs, Notion, or a simple text file) where
you store your most effective prompts. You can organize them by category, such as:

● Art Styles
● Specific Subjects
● Lighting and Composition
● Scenes and Environments

This way, you can quickly reference and copy-paste prompts as needed.

2. Use Prompt Templates

Build a set of reusable templates that you can modify for different needs. For example: "A
{subject} in a {setting} with {lighting} and {mood}, inspired by {artistic style}."

By keeping a collection of these templates, you can just fill in the blanks instead of typing
out detailed prompts from scratch.

3. Bookmark Prompts in Your Browser If you use a web-based

tool like DALL·E or MidJourney, save your most frequently used
prompts in a browser document or note-taking app, such as
Google Keep or Evernote. You can then quickly retrieve them
when needed.

4. Use Prompt Managers or Plugins

Some AI platforms or third-party apps allow you to store and manage your prompts directly.
For instance:
● Notion: Organize prompts and set up quick access links.
● Custom Prompt Managers: Store and retrieve your best-performing prompts, which is
useful for repeated use.
● If you're using a custom API or have access to developer tools, you might even be able
to create your own automated solution.

5. Automated Workflow with Shortcuts Use tools like text

expansion software (e.g., TextExpander or AutoHotkey) to
assign short keyboard shortcuts to frequently used prompts.
This way, typing a few characters will instantly expand into your
full prompt.

For example:

● Type “/cityscape” and it automatically pastes your predefined cityscape prompt

into the tool.

6. AI-Specific Features for Saved Prompts Some AI platforms

(like ChatGPT or DALL·E) may allow you to save custom
prompts within their interfaces. While this feature isn't available
on all platforms, check for any "favorites" or "saved prompts"
options that allow you to access your most used queries.

7. Create Custom AI Prompts in Your Local Tools If you're using

tools like Stable Diffusion or other locally run models, you can
create and save prompt files on your computer, then run them
directly whenever you need to generate images or results.

8. Prompt Repositories Libraries

Explore repositories that offer a collection of prompts, system prompts, and LLM
instructions. For example:
 The Big Prompt Library: A collection of various system prompts, custom instructions,
●
jailbreak prompts, GPT/instructions protection prompts, etc. It provides significant
educational value in learning about writing system prompts and creating custom GPTs1.
● Prompt Learning: Offers lessons on building prompt libraries, best practices for
maintaining and updating them, and collaborative prompt sharing.

9. Prompt Optimizers: Optimizing Prompts Across Multiple AI

Models Not all AI models will respond in the same way to identical prompts. For
example:
● ChatGPT-4 may be better at conversational depth, while Claude
excels at creative generation.
● Gemini Ultra may produce faster results but with a trade-off in
coherence.

Strategy: Compare how each model responds to a single prompt

and refine based on the model's strengths.

A) Using Embeddings and Vector Databases For highly

personalized interactions, you can use embeddings (dense
vector representations of text) and vector databases (e.g.,
Pinecone, Weaviate, FAISS) to store, retrieve, and fine-tune AI
responses. By embedding past interactions or domain-specific
knowledge, you can make your AI responses more tailored and
accurate.

B) RLHF (Reinforcement Learning from Human Feedback) RLHF

is a technique where AI models learn from human feedback to
improve their outputs over time. Using RLHF in prompting can
help create even more accurate, human-like results. For
advanced users, incorporating RLHF into your workflows can
make your prompts even more effective by constantly adapting
based on user corrections and feedback.

Use tools that help optimize your prompts for better performance:
● Quartzite AI: A free tool that helps you optimize your prompts by
applying best practices in prompt engineering.
● PromptPerfect: An AI prompt generator and optimizer that helps
you create, analyze, and optimize prompts.

10. Free Prompts Improvers

Utilize free tools to enhance your prompts:

● Galaxy AI: Offers a free AI prompt improver that transforms

basic prompts into powerful, detailed instructions.
● Prompthancer: A free AI-powered prompt enhancement
tool that improves your AI interactions.

11. Building Custom GPTs

To build custom GPTs and ease up prompting, follow these steps:

Write Clear and Direct Instructions: Break down multi-step

● instructions into simpler, manageable steps.

● Use “Trigger/Instruction Pairs”: Separate instructions with

delimiters to improve reliability.
● Promote Attention to Detail: Incorporate techniques like “take
your time” and “check your work” to ensure thoroughness.
● Avoid Negative Instructions: Frame instructions positively to
avoid confusion.
● Utilize Markdown and Structured Formatting: Enhance clarity
and readability with headings and structured formatting.

12. Advanced OpenAI Playground

Learn about advanced features of the OpenAI Playground to build

custom GPTs and Chatbots. This resource provides detailed
guidance on prompt engineering and creating effective GPTs9.
By utilizing one or more of these methods, you can save time and
avoid repetitive typing while maintaining the flexibility to modify
prompts as needed.

5.3 Comprehensive Guide to Building Custom

GPTs and Saving Prompts Efficiently
Creating custom GPTs (or fine-tuning an AI model) to ease up
prompting involves creating a system where users can interact with
specialized AI assistants that are tailored for specific tasks or
domains. These GPTs can understand the context with minimal
input, making prompt crafting easier and more efficient. Here’s how
to get started: 1. Define the Purpose and Use Case
● Identify Key Tasks: Determine what you want the GPT to
accomplish (e.g., generating product descriptions,
answering technical questions, providing creative
suggestions).
● Target Audience: Define who will use this GPT (e.g.,
business professionals, researchers, creatives, casual
users) to help shape the tone and complexity of the
prompts.

2. Design the Interaction Flow

 ● User Experience (UX): Ensure the interaction is smooth
and intuitive. Decide whether users will provide initial input
or if the GPT will ask guiding questions.
● Conversational Flow: Define how the GPT should
respond based on different inputs. For example, if a user
provides vague information, the GPT could ask clarifying
questions.
● Predefined Templates: Offer templates to streamline the
process. For example, a blog post template could guide
users through providing necessary context (topic,
audience, tone, length).

3. Train the GPT to Recognize Context

● Context-Aware Responses: Train the GPT to extract relevant
information from the user’s input and retain the conversation’s
context. For example, if a user asks for a technical guide, the
GPT should adjust its tone, depth, and level of detail accordingly.
● Contextual Reminders: Program the GPT to remind users about
any missing or unclear details. If a user is vague, the model can
ask for further context, such as clarifying terms, scope, or
intended outcome.

4. Fine-Tuning the Model (Optional)

● Custom Datasets: Gather a dataset of conversations or texts
specific to the domain you're targeting (e.g., customer service
scripts, legal documents, marketing content) to fine-tune the
model and make it more responsive.
● Example Input and Output Pairs: Provide sample conversations
or prompts that reflect real use cases to guide the GPT in
generating accurate responses.
● Pretrained Models: Leverage existing models (like GPT-3 or
GPT-4) and fine-tune them on a domain-specific dataset, saving
time compared to training a model from scratch.
When utilizing OpenAI's GPT-3 or GPT-4 models, it is customary to
interface with their API. The following provides a general outline of
the procedure: 1. Set Up Your Environment: Install necessary
libraries and obtain API keys.
import openai
openai.api_key = 'your-api-key'
2. Preprocess Your Text: Format your text samples in a way that's
optimal for the model. For example, you might break them into
smaller chunks.
def preprocess_text(text):
# Perform necessary text preprocessing here
return formatted_text
3. Fine-Tune the Model: Use the preprocessed text to fine-tune
your AI model. This involves feeding your text samples into the
model through a series of API calls.
def fine_tune_model(texts):
for text in texts:
response = openai.Completion.create(
model="text-davinci-003",
prompt=text,
max_tokens=1000
)
# Save or use the response as needed.
4. Test and Iterate: Once the model is fine-tuned, generate text
samples to see how well it mimics your style. Adjust parameters and
retrain as necessary.
test_prompt = "Write an introductory paragraph about climate change."
response = openai.Completion.create(
model="text-davinci-003",
prompt=test_prompt,
max_tokens=100
)
print(response.choices[0].text.strip()) 5. Incorporate Knowledge Bases and APIs
● Knowledge Base Integration: Link your GPT to a database or
external API (e.g., Google Scholar, medical databases, technical
documentation) for tasks like answering specific queries or
providing detailed insights.
● Retrieval-Augmented Generation (RAG): Implement RAG
models that can search for context online or in databases and
generate responses based on the gathered information, ensuring
up-to-date or domain-specific knowledge.

6. Create a Feedback Loop

● User Feedback: Collect feedback from users to identify
areas for improvement. For instance, after the AI provides
an output, prompt the user for feedback on whether it was
useful or what could be improved.
● Learning from Interaction: As your GPT interacts with
more users, it can learn common queries, preferred styles,
and frequent mistakes, refining its responses over time.

7. Build Error Handling and Clarification Prompts

● Fallback Responses: Ensure that if the GPT doesn’t fully
understand a user’s input, it asks for clarification or provides a
fallback suggestion to prevent confusion.
● Clarification Workflow: Develop a process where the GPT
checks if it has understood the user's context correctly. For
example:
○ User Input: "Can you explain quantum computing?"
○ GPT Response: "To make sure I understand your
request, would you like a general overview of
quantum computing or specific details like its
applications or mathematical foundations?"

8. Provide Post-Interaction Assistance

● Follow-Up Questions: After the initial response, the GPT can
suggest follow-up actions based on the initial query. For example,
if a user asks for a summary of an article, the GPT might suggest
deeper dives into specific sections based on the content’s
complexity.
● Further Refinement: Enable users to fine-tune the generated
results. Allow them to ask for additional modifications, such as
adjusting the tone, adding more examples, or elaborating on a
specific section.

9. Automated Context Expansion

● Automatic Prompting Assistance: A GPT can be built to
help users create detailed prompts themselves. For
example, if the user simply enters a topic, the GPT can
guide them to expand on it by asking questions like:
○ "What specific aspect of this topic would you like to
explore?"
○ "Do you have a target audience in mind?"
○ "What tone should the response take (formal,
conversational, humorous)?"
● Pre-Prompt Refinement: Develop an automated process
where users can specify if they want a short answer, a
detailed response, or even a specific type of content (e.g.,
list, essay, code snippet).

10. Saving and Reusing Prompts Efficiently

 Create a Prompt Repository: Keep a dedicated document or file
● (e.g., Google Docs, Notion, or a simple text file) where you store
your most effective prompts. Organize them by category, such as
art styles, specific subjects, lighting, and composition.
● Use Prompt Templates: Build a set of reusable templates that
you can modify for different needs. For example: "A {subject} in a
{setting} with {lighting} and {mood}, inspired by {artistic style}."
● Bookmark Prompts in Your Browser: Save your most
frequently used prompts in a browser document or note-taking
app, such as Google Keep or Evernote, for quick retrieval.
● Use Prompt Managers or Plugins: Tools like Notion, custom
prompt managers, or APIs can store and retrieve your best-
performing prompts.
● Automated Workflow with Shortcuts: Use text expansion
software (e.g., TextExpander or AutoHotkey) to assign short
keyboard shortcuts to frequently used prompts.
● AI-Specific Features for Saved Prompts: Some AI platforms
(like ChatGPT or DALL·E) may allow you to save custom prompts
within their interfaces. Check for any "favorites" or "saved
prompts" options.
● Create Custom AI Prompts in Your Local Tools: For tools like
Stable Diffusion or other locally run models, create and save
prompt files on your computer for direct use.

11. Explore Prompt Repositories and Optimizers

Prompt Repositories Libraries:
●

○ The Big Prompt Library: A collection of various

system prompts, custom instructions, and LLM
instructions.
○ Prompt Learning: Lessons on building prompt
libraries, best practices, and collaborative sharing.
● Prompt Optimizers:
 ○ neuroflash: Free tool that optimizes prompts using
best practices in prompt engineering.
○ PromptPerfect: An AI prompt generator and
optimizer. Visit PromptPerfect
● Free Prompts Improvers:
○ Galaxy AI: Free prompt improver for transforming
basic prompts into powerful instructions.
○ Prompthancer: AI-powered tool for enhancing
prompts.

6. Outro: Unlocking Your Prompting

Potential
Thank you for joining us on this journey into the fascinating world of
prompting. Mastering the art and science of prompting is more than
just a skill—it’s a gateway to leveraging the full potential of AI,
unlocking creativity, efficiency, and innovation across every aspect of
your personal and professional life.

Download More Resources: Our journey doesn’t end here. To

support your continued growth as a prompt master, we’ve
collected a wealth of resources to help you hone your skills:
● Anthropic's Prompt Engineering Interactive Tutorial: This
course is intended to provide you with a comprehensive step-by-
step understanding of how to engineer optimal prompts within
Claude.
● Marketing Prompt Engineering: Ready-to-use prompts for
marketing. Common basics of prompt engineering and examples.
● Generative AI Advanced Guide: Learn how you can seamlessly
integrate generative AI into your company or your day-to-day
work.

Where to Learn More:

The prompting community thrives on shared knowledge and
collaboration. Have you created a particularly effective prompt?
Solved a challenging problem using an AI tool? We’d love to hear
about your successes!

● Community Forums: Join discussions with fellow learners

and professionals to exchange ideas and refine your skills.
Like The DeepLearning Community Forum, or The Dair-
AI’s Discord Channel.
● Workshops and Webinars: Check out this Google's
webinar where experts break down essential prompting
techniques and there’s zero experience required. Or this
ChatGPT Prompt Engineering for Developers in case you
are already a pro at prompting engineering.

Looking Ahead

As you step into the world of prompting, remember that this is just
the beginning. The skills you’ve developed are not only essential
today but will continue to evolve as AI becomes an even more
integral part of our lives. Whether you’re solving complex problems,
creating stunning visuals, or driving innovation in your industry, the
ability to communicate effectively with AI will remain one of the most
valuable tools in your arsenal.

Looking forward, it becomes clear that the art of prompting is

destined for transformation. The techniques we employ today will
grow more sophisticated, embracing greater nuance and
adaptability. Imagine a future where user-friendly interfaces and
advanced LLMs collaborate seamlessly, reducing the need for
explicit instructions. Contextual understanding will deepen, allowing
AI to intuitively grasp the essence of our queries and enabling a new
era of effortless interaction.

Together, we’re shaping the future of human-AI collaboration. By

refining our prompting techniques and exploring the ever-expanding
capabilities of AI, we pave the way for transformative innovation.
Let’s continue to explore, innovate, and redefine the way we interact
with technology.

Thank you for being part of this exciting journey. Here’s to your
success as a prompt master and to the limitless possibilities that lie
ahead!
More Resources:
● Best Practices for Prompt Engineering with the OpenAI API.
● Prompts to humanize AI Response Using GPT-4.
● Stable Diffusion Camera Distance and Angles - Controlling
camera distance and angles.
● The Ultimate Stable Diffusion Camera Angle Prompts Guide - A
comprehensive guide on controlling camera angles.
● SDXL Guide to Camera Framing and Angle - A guide on
controlling the view for XL models
● Discover How Camera Positions Can Elevate Your AI Art and
Generative AI Storytelling
● AI Image Generation Prompt Examples
● Community OpenAI: DALL·E-3 Prompt Tips and Tricks
● The DALL·E-2-prompt book
● Stable Disffusion Prompts
● 2023 AI Index Report

● 3 Principles for prompt engineering with GPT-3

● Eight Things to Know about Large Language Models
● A beginner-friendly guide to generative language models -
LaMBDA guide
● A Complete Introduction to Prompt Engineering for Large
Language Models
● A Generic Framework for ChatGPT Prompt Engineering
An SEO’s guide to ChatGPT prompts
●

● Anyone can Design! With a little help from Generative AI

● AI Content Generation
● AI's rise generates new job title: Prompt engineer
● AI Safety, RLHF, and Self-Supervision - Jared Kaplan | Stanford
MLSys #79
● Awesome Textual Instruction Learning Papers
● Awesome ChatGPT Prompts
 Awesome ChatGPT Prompts
● Best 100+ Stable Diffusion Prompts
● Best practices for prompt engineering with OpenAI API
● Building GPT-3 applications — beyond the prompt
● Can AI really be protected from text-based attacks?
● ChatGPT, AI and GPT-3 Apps and use cases
● ChatGPT Prompts
● ChatGPT3 Prompt Engineering
● CMU Advanced NLP 2022: Prompting
● Common Sense as Dark Matter - Yejin Choi | Stanford MLSys #78
● Create images with your words – Bing Image Creator comes to
the new Bing
● Curtis64's set of prompt gists
● CS324 - Large Language Models
● CS 324 - Advances in Foundation Models
CS224N: Natural Language Processing with Deep Learning
●

● DALL·E 2 Prompt Engineering Guide

● DALL·E 2 Preview - Risks and Limitations
● DALLE Prompt Book
● DALL-E, Make Me Another Picasso, Please
● Diffusion Models: A Practical Guide
● Exploiting GPT-3 Prompts
● Exploring Prompt Injection Attacks
● Extrapolating to Unnatural Language Processing with GPT-3's In-
context Learning: The Good, the Bad, and the Mysterious
● FVQA 2.0: Introducing Adversarial Samples into Fact-based Visual
Question Answering
● Generative AI with Cohere: Part 1 - Model Prompting
● Generative AI: Perspectives from Stanford HAI
● Get a Load of This New Job: "Prompt Engineers" Who Act as
Psychologists to AI Chatbots
● Giving GPT-3 a Turing Test
● GPT-3 & Beyond
● GPT3 and Prompts: A quick primer
● GPT-4 Tutorial: How to Chat With Multiple PDF Files (~1000
pages of Tesla's 10-K Annual Reports)
● Hands-on with Bing’s new ChatGPT-like features
● How to Draw Anything
● How to get images that don't suck
How to make LLMs say true things
●

● How to perfect your prompt writing for AI generators

● How to write good prompts
● If I Was Starting Prompt Engineering in 2023: My 8 Insider Tips
● Indirect Prompt Injection on Bing Chat
● Interactive guide to GPT-3 prompt parameters
● Introduction to ChatGPT
● Introduction to Reinforcement Learning with Human Feedback
● In defense of prompt engineering
● JailBreaking ChatGPT: Everything You Need to Know
● Long Context Prompting for Claude 2.1
● Language Models and Prompt Engineering: Systematic Survey of
Prompting Methods in NLP
● Language Model Behavior: A Comprehensive Survey
● Learning Prompt
● LINGO : Visually Debiasing Natural Language Instructions to
Support Task Diversity
● Make PowerPoint presentations with ChatGPT
● Meet Claude: Anthropic’s Rival to ChatGPT
 Meet Claude: Anthropic s Rival to ChatGPT
● Methods of prompt programming
● Mysteries of mode collapse
● NLP for Text-to-Image Generators: Prompt Analysis
● NLP with Deep Learning CS224N/Ling284 - Lecture 11:
Prompting, Instruction Tuning, and RLHF
Notes for Prompt Engineering by sw-yx
●

● On pitfalls (and advantages) of sophisticated large language

models
● OpenAI Cookbook
● OpenAI Prompt Examples for several applications
● Pretrain, Prompt, Predict - A New Paradigm for NLP
● Prompt Engineer: Tech's hottest job title?
● Prompt Engineering by Lilian Weng
● Prompt Engineering 101 - Introduction and resources
● Prompt Engineering 201: Advanced prompt engineering and
toolkits
● Prompt Engineering 101: Autocomplete, Zero-shot, One-shot, and
Few-shot prompting
● Prompt Engineering 101
● Prompt Engineering - A new profession
● Prompt Engineering by Microsoft
● Prompt Engineering: The Career of Future
● Prompt Engineering Guide: How to Engineer the Perfect Prompts
● Prompt Engineering in GPT-3
● Prompt Engineering Template
● Prompt Engineering Topic by GitHub
● Prompt Engineering: The Ultimate Guide 2023 [GPT-3 & ChatGPT]
● Prompt Engineering: From Words to Art
 Prompt Engineering with OpenAI's GPT-3 and other LLMs
●

● Prompt injection attacks against GPT-3

● Prompt injection to read out the secret OpenAI API key
● Prompting: Better Ways of Using Language Models for NLP Tasks
● Prompting for Few-shot Learning
● Prompting in NLP: Prompt-based zero-shot learning
● Prompting Methods with Language Models and Their
Applications to Weak Supervision
● Prompts as Programming by Gwern
● Prompts for communicators using the new AI-powered Bing
● Reverse Prompt Engineering for Fun and (no) Profit
● Retrieving Multimodal Information for Augmented Generation: A
Survey
● So you want to be a prompt engineer: Critical careers of the future
● Simulators
● Start with an Instruction
● Talking to machines: prompt engineering & injection
● Tech’s hottest new job: AI whisperer. No coding required
● The Book - Fed Honeypot
● The ChatGPT Prompt Book
The ChatGPT list of lists: A collection of 3000+ prompts,
●
examples, use-cases, tools, APIs, extensions, fails and other
resources
● The Most Important Job Skill of This Century
● The Mirror of Language
● The Waluigi Effect (mega-post)
● Thoughts and impressions of AI-assisted search from Bing
● Unleash Your Creativity with Generative AI: Learn How to Build
Innovative Products!
● Unlocking Creativity with Prompt Engineering
● Using GPT-Eliezer against ChatGPT Jailbreaking
● What Is ChatGPT Doing … and Why Does It Work?
Academic Papers:

● Personalizing Text-to-Image Generation via Aesthetic Gradients -

Research on personalizing text-to-image models.
● An Image is Worth One Word: Personalizing Text-to-Image
Generation using Textual Inversion - Study on Textual Inversion
technique.
● The Prompt Report: A Systematic Survey of Prompting

Techniques (June 2024)

● Prompt Design and Engineering: Introduction and Advanced
Methods (January 2024)
● A Survey on Hallucination in Large Language Models: Principles,

Taxonomy, Challenges, and Open Questions (November 2023)

An RL Perspective on RLHF, Prompting, and Beyond (October
●
2023)
● Few-shot Fine-tuning vs. In-context Learning: A Fair Comparison

and Evaluation (May 2023)

● Jailbreaking ChatGPT via Prompt Engineering: An Empirical Study
(May 2023)
● Harnessing the Power of LLMs in Practice: A Survey on ChatGPT
and Beyond (April 2023)
● Tool Learning with Foundation Models (April 2023)
● One Small Step for Generative AI, One Giant Leap for AGI: A
Complete Survey on ChatGPT in AIGC Era (April 2023)
● A Bibliometric Review of Large Language Models Research from
2017 to 2023 (April 2023)