Prompt engineering is quickly becoming one of the most valuable skills for working with AI, and this video breaks it down in a way that's simple to follow even if you've never written a prompt before. You'll learn what large language models are, how tools like ChatGPT, Gemini, and Claude actually work behind the scenes, and why some prompts get far better results than others. From understanding prompt structure to applying best practices for clear, effective prompts, this lesson gives you a solid foundation you can start using right away, whether you're drafting content, solving problems at work, or exploring AI for the first time. It's designed for business professionals, students, content creators, and anyone curious about artificial intelligence who wants practical, real-world skills rather than technical jargon. No coding background needed, just curiosity and a willingness to learn. Stick around, because understanding how these AI systems think is the first step to getting real, useful results from them.
🎬 TIMESTAMPS
SECTION 1: INTRODUCTION 0:00:00 Introduction 0:00:33 Meet Your Instructor 0:01:21 What is Artificial Intelligence? 0:03:46 Introduction to Prompt Engineering
SECTION 2: LARGE LANGUAGE MODELS (LLMs) 0:06:04 What are Large Language Models? 0:07:06 How Do Language Models Work? 0:09:01 Popular Examples of LLMs 0:09:57 Strengths & Limitations of LLMs 0:11:29 What is Hallucination? 0:12:43 How to Use ChatGPT 0:14:17 How to Use Gemini by Google 0:15:42 Claude & Copilot 0:17:07 Meta AI (WhatsApp AI) 0:17:57 Free vs Paid Versions
SECTION 3: PROMPTING FUNDAMENTALS 0:18:58 Prompt: Meaning & Types 0:20:14 Prompt Structure 0:20:53 Best Practices for Writing Prompts
SECTION 4: PROMPT ENGINEERING TECHNIQUES 0:21:57 How to Develop Effective Prompts 0:23:08 Few-Shot Prompting 0:27:40 Guardrails 0:28:42 Meta Prompts 0:34:24 Chain of Thought (CoT) 0:36:44 Prompt Chaining vs Prompt Chunking 0:47:01 Summary of Prompt Engineering Techniques
SECTION 5: AI SECURITY & SAFETY 0:47:01 Public Nature of LLMs 0:49:11 Sharing Sensitive Information 0:50:34 Ethical & Responsible Use 0:52:52 Limitations of AI Privacy
SECTION 6: ADDITIONAL LEARNING & ASSESSMENT 0:55:12 Multi-Step Prompting 0:57:34 Assessment & Example 0:59:48 Conclusion 1:00:23 End of Course
AI Generated·May contain errors
By now, you may have already heard about Artificial Intelligence or AI chatbots.
You may have also heard of ChargPT, Gemini, Copilot, Perplexity or Mistral, and many other AI tools.
In this course, you will learn how to get the most out of AI chatbots by asking the right questions
in the right manner to get the right results. This process is also known as font engineering.
Hello there. Welcome to the Prompt Engineering for Beginners course. My name is Filippa Diakombi
and I am the founder and CEO of Decision Spark Technologies and one of the pioneer AI product
leaders in Africa. Vertical has partnered with Decision Spark Technologies to bring this course
to you. In this course, I'll guide you through the basics of how AI chatbots and large language
models work and how to create prompts that help them perform better. You'll learn about
different types of prompts, strategies to improve responses, and we'll learn some advanced
techniques. Whether you're using AI for content creation, coding, or problem solving,
this course will provide the tools to enhance your skills.
What is Artificial Intelligence, otherwise known as AI? Artificial Intelligence is the mechanism
of how machines, particularly computer systems, are programmed to exhibit human-like capabilities
such as learning from data, applying reasoning to solve problems, or correcting their own errors.
The key word here is data. Data is all around us. The time you woke up this morning,
the type of mobile phone you used, the number of cars on the street, your favorite songs,
your favorite videos, and many more are all classified as data you've collected.
These are also called data points. How are these computer systems trained? It is a process called
machine learning. As humans, we learned all of what we can do today in a process called
innate or acquired learning. And like humans, computer systems are taught to identify patterns
from data, and this process is known as machine learning. A machine is learning. This means
computer systems learn from data in a process called machine learning to become artificially
intelligent. So, data, machine learning, and artificial intelligence. Every chatbot that
interacts with today has all been trained on a large data set of internet texts, which are
billions of data points, and that is why some chatbots tell you that they cannot provide answers
from a particular year or they make up something incorrect instead. This is because chatbots were
not trained on the specific data to that year. For example, when Chargibility launched in 2022,
it couldn't answer questions from several years ago. So, when a response from a chatbot is audited,
it could be due to the training data used. The majority of the creators of these chatbots and AI
tools continue to update their tools with more data points. These chatbots belong to a family
known as conversational AI. This means they have been trained to give output in a conversational
manner, making it easy for the punter to understand the output. AI chatbots are great companions for
professionals and students alike. AI chatbots are a means to an end, but not the end itself.
Think of a calculator. A calculator doesn't take away your math skills, but it assists you to
achieve your mathematical results faster. Effectively using AI chatbots help to generate new ideas,
grow existing ideas, lighten your workload from writing emails, generating code, to image designs.
Now, let's talk about prompt engineering. In this module, we'll talk about prompt engineering,
the importance of prompt engineering, and a brief history of language models.
Now, let's get started. Prompt engineering is the process of designing and optimizing text prompts
to elicit specific responses from language models housed by chatbots. It is the process of asking
questions in a way that gets the clearest, most useful answer. Just like how you might
phrase a question differently depending on who you're talking to, with AI, how you frame a prompt
can greatly impact the response you get. It is the art of crafting precise instructions or prompts
to guide AI models, like large language models, LLMs, to generate the desired output.
So, what are the importance of prompt engineering? Prompt engineering is highly important because
while AI chatbots are powerful, they rely heavily on how we guide them. A well-crafted prompt can
result in accurate, detailed, relevant, and quality responses. On the other hand,
a poorly designed prompt might lead to confusion, vague, or even incorrect answers.
By learning prompt engineering, you will know how to unlock the full potential of
language models behind AI chatbots and use it effectively in writing, coding, research,
idea generation, and many more. Let's talk about the brief history of language models.
A language model is a machine learning model that uses statistics and probabilities to predict and
generate language. Every AI chatbot is powered by language models. The history of language models
spans decades, but the leap forward came in recent years with the development of transformers,
the architecture behind models like GPT, the models behind chatGPT. Over the years,
language models have evolved rapidly. With the development of large language models, LLMs,
like OpenAI's GPT, Meta's LAMMA, Google's BERT, and Lambda's UTC, AI chatbots can now understand
and generate human-like text. LLMs are trained on huge amounts of data and learn to predict and
generate text that feels natural and coherent. Understanding AI and large language models.
In this module, we'll talk about LLMs, the types of LLMs,
how LLMs work, and the strengths and limitations of LLMs. Let's dive right in.
What are LLMs? Large language models, otherwise known as LLMs, are AI systems designed to
understand and generate text in a way that mimics human language. They are built using
the technique called deep learning and trained on vast amounts of text data. This allows them
to perform a wide variety of language user tasks, from answering questions to writing essays,
emails, promotional materials, or holding various forms of conversations.
How do language models work? LLMs work by predicting what comes next
in a sequence of words based on the input given. As previously noted, LLMs are trained
on massive data sets from the internet, books, and other sources. The more data and training
they receive, the better they become at understanding the nuances of language.
However, they are not perfect. They don't actually understand text like humans do. Instead,
they are really good at pattern recognition and producing text that statistically fits
the context of a given prompt. Think of an LLM as a massive interconnected network.
It processes input text, analyzes patterns, and generates outputs based on its understanding of
language and context. Watch this video from Mira Murati, the former CTO of OpenAI, the creator
of ChartGPT. This is on publicly available data that was used to train Sora, OpenAI's text-to-video
model. Different types of LLMs are available publicly today, each with their unique strength.
Popular examples include GPT, generative pre-trained transformers, used in models like
ChartGPT. BERT, bidirectional encoder representations from transformers,
specialized in understanding context by reading text in both directions. Another popular example
is TIFI, text-to-text transfer transformer, which converts all tasks into text-to-text formats
like translation, summarization, etc. A popular one also is PALM, pathway language model. This
works by learning to represent relationships between phrases and words in sentences.
What are the strengths and limitations of LLMs? LLMs are incredibly powerful and can understand
and handle a wide range of tasks, but they have limitations. What are the strengths? They can
generate creative content, assist with coding, answer factual questions, and simulate conversations.
Example of commonly used AI transport. OpenAI's ChartGPT. This is great for conversational tasks.
Google's Gemini. This is very good for understanding context in search queries.
OpenAI's Codex. This specializes in generating code. Meta's LLAMP specializes in generating
code and understanding complex programming tasks. Then you have Microsoft's Copilot,
ideal for integrating AI assistance into productivity tools like Word,
Excel, PowerPoint, helping users to write, analyze data, and streamline tasks.
So what are some of the limitations? It can sometimes produce incorrect information,
fail to understand complex tasks, or generate irrelevant responses. This is known as
hallucination. Also, they don't have reward awareness. They only know what they've been
trained on. Let's do a quick refresher on hallucination. Hallucination occurs very often
during conversations with AI transports. Hence, it is very important for front engineers to pay
close attention to this. Hallucination refers to instances where an AI model generates information
that seems plausible, but it's actually incorrect or fabricated. This happens because the model
doesn't know facts. It only predicts the most likely words based on patterns it has learned.
For instance, if you ask a model to generate a historical fact it doesn't know,
it might make up an answer. What are the solutions to hallucination? You have to verify
information. Cross-check the AI's response with trusted sources. You have to provide
constraints. If you want factual information, ask the chatbot to use only verified data or
avoid speculating. You have to also use guardrails. Be specific in your prompt if you are expecting
factual, verifiable responses, like give only documented facts in your prompt.
Now that we've explored what LLMs are and how they work, let's shift from theory to practice.
We'll learn how to start using some of the popular LLM tools available today,
even if you've never used one before. By the end of this lesson, you'll learn how to sign up,
log in, and start chatting with AI tools like ChargeBT, Gemini, Cloud, Copilot, and even Meta
AI on WhatsApp. Let's go step by step. Let's start with ChargeBT by OpenAI. What is it? ChargeBT is
one of the most popular conversational AI chatbots. It's available as a website, a mobile app,
and even a browser extension. How do you sign up? Visit chat.openai.com, click on sign up,
you can register with your email address, a Google account, or a Microsoft account. You may
be asked to verify your phone number for security, and it comes in a free and paid plan.
Where to access it? On desktop, go to chat.openai.com. On mobile, download the ChargeBT app
from your Google Play Store or Apple Store. On browser, use extensions like ChargeBT for Google
to get answers alongside your search results. Pro tip, start with the free version. Upgrade
only if you need advanced features like file uploads, or faster performance, or image generation.
Secondly, let's talk about Gemini by Google, formerly known as Bad AI. Gemini is Google's
conversational AI that integrates deeply with Google's search and other Google tools.
How to sign up? Go to gemini.google.com, click on sign in using your Google account,
and phone numbers may be needed, but if you already have a Google account, it's not needed.
There's also a free and paid plan for Gemini. The free plan includes basic Gemini access,
while the paid plan is available through Google One AI Premium, which gives you access to Gemini
advanced features. Where to access it? On desktop, use your browser to go to gemini.google.com.
On mobile, you can access Gemini through Google Apps or Android devices directly. Gemini may
replace Google Assistant on some phones one day. On integration, Gemini is now embedded in Gmail,
Google Docs, and Google Sheets, if you're already using Workspace tools. Try asking Gemini questions
directly related to search like, what's the difference between a visa and a resident permit
in Canada? Thirdly, Cloud by Anthropic. Cloud is another safe conversational AI that focuses on
ethical alignment and reasoning. It's known for long memory and handling complex documents. How
to access Cloud? Visit cloud.ai, sign up with your email, currently free in supported regions,
with paid options coming soon as announced by them. Next is Microsoft Copilot. Copilot is
Microsoft's AI assistant built into tools like Word, Excel, Outlook, Edge, which are Microsoft's
products. It is powered by OpenAI models, and how do you use it? In Microsoft 365, Copilot appears
directly in Word, Excel, and Teams for paid enterprise accounts. On Windows 11, look for
the Copilot icon on your taskbar. On Edge browser, click on the Copilot sidebar on the right.
How do you access it? For personal use, visit copilot.microsoft.com. For enterprise users,
this is enabled through Microsoft 365 Copilot licenses. Some tips, use it to draft emails in
Outlook, generate summaries in Word, or explain formulas in Excel.
The next one, MetaAI, the popular WhatsApp AI assistant. MetaAI is now integrated into WhatsApp,
Facebook, Messenger, and Instagram DMs. It works like a chatbot that can help you draft messages,
answer questions, and generate content. It is completely free to use. How to use it on WhatsApp?
Make sure you have the latest version of WhatsApp. In some regions, simply start a new chat with
MetaAI. You can also type at MetaAI in a group chat to interact with it. A pro tip, ask it to
help you with better messaging ideas, translations, and general questions, just like you would ask
Google. Here are some general tips when choosing between free and paid plans. Start free. All
platforms offer free versions that are powerful enough for your everyday use. You can upgrade
later when you need features like file uploads, advanced reasoning, faster response time, priority
access during peak hours, image generation, audio generation, video generations. Compare features.
Check side-by-side differences before subscribing.
Fundamentals of prompting. In this module, we'll talk about prompts and its types,
structures, examples, and best practices for writing prompts. Without further ado, let's dive
right in. What is a prompt? A prompt is simply the input or instruction you give to a language model
to get a response. It can be a question, a sentence, or even just a few words. The quality
of the prompt largely determines the quality of the output. What are some of the prompt types?
Open-ended prompts, closed-ended prompts, conditional prompts. What are open-ended prompts?
These encourage creative or detailed responses. An example, write a short story about an intercity
road trip. Closed-ended prompts. Let's talk about an example as well. These expect specific and
concise answers. For example, who is the seventh director-general of the World Trade Organization?
The answer you would get is Ngozi Okonjo-Iweala. Or, what is the capital of France? The answer
expected is Paris. Conditional prompts. These depend on prior contact information. For example,
we can say, based on the previous answer, explain why Paris is the capital of France.
Let's talk about prompt structure. A good prompt typically has three parts. The input,
the output, and the context. The input is the information you provide. The output is the result
you expect. The context is any background information needed to clarify the task.
Examples. Based on the context of the passage, explain the main arguments in 200 words.
You can also say, using the information provided, summarize the key contents through a chatbot.
Best practices for writing prompts. You must ensure you are clear,
and you use concise sentences. You must provide enough context to avoid confusion.
You must use examples to guide the AI toward the response you want. For example, instead of saying,
explain photosynthesis, try, explain the process of photosynthesis in simple terms
for a class with seven-year-olds.
Prompt engineering techniques. How to develop effective prompts. In this module,
you will learn about structure of language, avoiding ambiguity, prompt chaining,
Qshort prompting, chain of thought prompting, and other prompt engineering techniques.
Let's dive right in.
Structure of language, specificity, and avoiding ambiguity.
When creating prompts, it is crucial to understand the structure of language.
Specifying what you want is highly important. Vague or ambiguous prompts can lead to poor
results, as we've already mentioned. For example, instead of saying, tell me about history,
you could say, explain the significance of the French Revolution in world history.
Another example, instead of asking, write something about a cat, you could ask,
write a short story about a cat who dreams of flying to the moon. This ensures the model
knows exactly what you're looking for, and this expansion helps the output.
Providing examples can help the LLM or AI chatbot to understand the desired output.
For instance, if you want the model to write a poem in a specific style, you can provide
a few examples of the poem in that style, and you can tell the chatbot to write you an email,
like the example you provided.
Let's talk about the different prompt engineering techniques. First,
let's talk about few-shot prompting. Few-shot prompting is a technique of giving the AI
chatbot one or more examples of the kind of answer expected before asking the actual question. This
way, you can guide the LLM powering the AI chatbot behind to learn quickly and give you a desired
response or output. Let's take this, for example, example one, movie title, Inception.
You are creating a sermon to say, a skilled thief who enters the dreams of others to steal
secrets is tasked with planting an idea in someone's mind, leading to a complex journey
through layers of dreams. You can also give it another example with the movie Titanic,
which the summary says, a young couple from different social backgrounds fall in love
about the ill-fated RMS Titanic, which then sinks after hitting an iceberg.
Now you tell the chatbot to say, now summarize the movie, the matrix, and you will see that
the response it gives you is in line with the examples from Inception and Titanic.
So let's talk about few-shot prompting and some examples. This example that you're looking at
is for an entrepreneur looking to generate business ideas for expansion of his delivery
business. So I am prompting the chatbot, this is chatGPT, and I'm asking it to say,
you are the founder of a successful food delivery business looking to expand into new markets
or deliver new services. Suggest a new business line for a healthy meal delivery.
On this page, you can see that the prompt is just basically asking the chatbot to
bring up suggestions based on this alone. Let's see what the response is. As you can see,
it starts generating, and it says, as the founder of a successful food delivery business
focused on healthy meals, expanding into new markets or services can help further differentiate
your brand and tap into new customer needs. And it gives suggestions. You can see it's talking
about personalized nutrition plans and meal kits. You see the overview, you see the key components
for this exact plan. You also see the meal kits for home cooking. You see the on-demand coaching
as well. And these are ideas that the founder can expand upon or tweak regarding the services
that the founder looks to offer. You can see also an idea that it has given to say,
to give a subscription model. And this includes offering flexible subscription options for
weekly, bi-weekly, or monthly plan so that the customers can commit to the service for as long
as they need while accommodating their lifestyle. So you see these ideas that the chatbot now starts
to give you versus just trying to get the information without the right prompting.
You also see an example of it giving a suggestion of integrating with fitness apps. So this way,
the services are then expanded. As you can see, it's also talking about why this would work.
So you see personalization, you see convenience, you see health and wellness, you see scalability
as well. And it gives an example. You could consider expanding into corporate wellness
programs by offering these personalized services to companies as part of your employee wellness
packages and helping employees manage their health and nutrition during busy workdays.
So you see the different examples it starts to give just based on the prompt that was given.
So let's talk about guardrails as well. Using guardrails is a technique that can be used to
set boundaries on the type of desired responses. For example, instead of leaving a prompt open
ended, you can provide an answer in no more than 100 words. And this guides the AI chatbot to give
you a response in that scope. So for example, you can say to the chatbot, write a creative story
about a Nigerian superhero who gains power from eating jellyfries. However, the story must adhere
to the following guidelines. No explicit violence or positive portrayal of Nigerian culture and
traditions. The superhero's power should be realistic and grounded in the concept of good
energy. Another guardrail you can use is the story should promote positive values like courage,
honesty, and possible rights. Now, metaprompts. As with guardrails I mentioned previously,
metaprompts are also a technique that can be used to tell the LLMs how to behave to engineer
a desired response. For instance, you can say act as a history professor and explain the topic.
In this example, as you have on your screen, you can say you're a financial advisor with 10 years
of experience. Provide a detailed financial plan for someone who wants to start saving
for retirement in 20 years. Assuming they are currently 30 years old, they earn $50,000 annually
and have minimal savings. Explain the steps they should take, including investment strategies,
saving goals, and potential risks. This way, the chatbot provides a response that
takes in all these prompts and gives you the desired response.
Let's take an example. Let's take an example for a metaprompt for a marketing and communications
executive for strategy and planning, and we are using chatGPT once again. In the prompt,
we are saying to the bot, act as a marketing and communications executive and develop a
marketing plan to generate a six-month marketing strategy for launching a new product line for
a hiring and talent management firm. The strategy should include budgets, allocation,
and key performance indicators, KPIs. Let's see what the response is. As we can see,
it starts with an overview, and it gives an overview of the entire business in hiring and
talent management. You can also see it saying that you can use tools like ATS,
equipment platforms, and others based on the prompts that it has been given.
It gives you your marketing objectives to build awareness, your lead generation,
your market penetration, and ideas around customer acquisition and retention.
You can see also it's talking about the target audience, who are your primary target audience,
or your secondary target audience. You also see ideas around key messaging, and you can make all
the strategy work for you. You can even tweak your prompts however you want to give you your
desired results. Here you see the marketing channels, and this helps you to know where to
launch your marketing efforts. You see your digital marketing, your email marketing,
your social media marketing, and the different tactics and timelines, and even budgets that
you can use. And all this, once again, can be tweaked by you if you are a media communications
executive. So you see also using influencer partnerships and PR, page search and display ads,
using Google search, Google ads, and the likes. You also see offline marketing as an idea that
Chatspot has given for industry events and conferences, sponsoring and exhibiting at HR
conferences, trade shows and webinars, organizing product demos and face-to-face
consultations with prospects. It also shows you a response regarding direct mails.
After all that has been done, it now shows you budgets for your marketing efforts, and you can
see right there what the budget looks like for your marketing efforts, and it gives you a table
for every cost for your marketing efforts. You also then see the key performance indicators,
what you can measure your PR initiatives on, or your marketing and communications initiatives on.
So you see your KPIs for your brand awareness, your lead generation, your sales and revenue,
customer feedback and satisfaction. Once again, all this can be tweaked however you want
as a marketing and communications executive. You also see a response regarding timeline and
execution plans, and it gives you a table that shows a month-on-month execution plan and the
activities that you can use to bring your objectives to life. So you see for month one,
it's telling you how to make your products launch up on your content development, your page search
and influencer outreach. You see for your month two, the activities you need to do,
the activities you need to do for month three, month four, and month five, and up to month six,
right? And all these are based on the prompts that was given to the chatbots, right? And that's
an example of a meta prompt for a marketing and communications executive. Another interesting
response the chatbot gives is a post-launch activities. So you see the chatbots telling
you what a post-launch activities should look like, right? From customer onboarding and support
to serving customers, to reviewing and refining your strategies. Once again, you can take this,
work on it. You can refine it however you want. You can tweak however you want and use for your
success. Now, let's talk about some interesting concepts. Chain of thoughts, COT prompting. Chain
of thoughts prompting is a technique that improves the performance of LLMs by guiding them to break
down complex tasks into a series of steps. This technique offers significant benefits in various
domains, such as improved accuracy, transparency, and multi-step reasonability. Let's take this
example. So we are looking at standard prompting versus a chain of thoughts prompting. In the
standard prompting, we see the model input saying, Roger has five tennis balls. He buys two more cans
of tennis balls. Each has three tennis balls. How many tennis balls does he have now? The answer is
11. Then a follow-up question to say, the cafeteria has 23 apples. If they use 20 apples to make lunch
and bought six more, how many apples do they have? The model output says 10 times 27,
which isn't correct. And if you look at the example for the chain of thoughts prompting,
it follows a layered approach and it gives a response that is accurate. So let's talk about
ZeroShot's chain of thoughts prompting, COT. This is a form of chain of thoughts prompting that
involves a more narrowed-down approach of prompting. Adding things step-by-step to the original prompt
really does the wonders. In this example, the prompt says, I went to the boutique and bought
10 shirts. I gave two shirts to my neighbor and two to my friend. I then went and bought five more
shirts and I took one. How many shirts am I left with? And this is where you add ZeroShot's chain
of thoughts prompting. You say, think step-by-step, and you can see the output telling you, let's
break down the problem step-by-step. And it gives you the initial purchase, you bought 10 shirts.
Gifts, you gave away 10. You gave away two, plus two, meaning you gave away four shirts.
It tells you the second purchase, that you bought five more shirts. Then for personal use, you took
one shirt for yourself. So you can see how it breaks down this problem in a step-by-step manner
to give a response, which eventually tells you that you're left with 10 shirts.
Now let's talk about prompt chaining. Prompt chaining involves using a series of prompts to
lead the model through a process. It's a technique of breaking down complex tasks into smaller,
interconnected prompts. For example, you can ask the AI chatbot to summarize a text, then ask it
to explain a specific part, or tell an AI chatbot to write a story using a sequence of prompts.
In this example, we see how the prompts elicit a response. For prompt one, you can say,
write a short story about a young Nigerian entrepreneur who starts a tech startup in Lagos.
Then for prompt two, in the story, introduce a challenge the entrepreneur faces due to the
unreliable power supply in Nigeria. Then for prompt three, you can say, now write a scene
where the entrepreneur pitches the idea to a potential investor, highlighting how they plan
to overcome the challenge of power outages. And you can see the sequence of prompting would enable
a response using prompt chaining, and you can see the response that the chatbot gave.
Now we get to prompt chunking. Let's talk about prompt chunking. Prompt chunking is a technique
to break down complex prompts into smaller, more manageable chunks. This helps to improve the
clarity and focus of the prompt, making it easier for daily learners to understand and process.
It is especially useful when dealing with complex queries or large volumes of information.
For example, you can write a prompt that says, design a weekly menu for a restaurant.
Note the following below. You give it lists of appetizers, main courses, desserts, beverages,
and specials of the day, and you see how that really helps the chatbot to give a response
aligned to what you want. So prompt chunking and prompt chaining seem similar, but let's talk
about the differences. So for prompt chunking, let's look at the differences between prompt
chunking and prompt chaining within focus, goals, and output. So in terms of focus,
prompt chunking will break down a single complex prompt into smaller, simpler prompts to provide
clarity and focus. With prompt chaining, there's a sequence of prompts created where the output
of one prompt becomes the input of the next. In terms of goals, prompt chunking's goals are to
make the prompt more understandable for the language model and guide it through and towards
a specific output. For prompt chaining, the goal is to generate a series of outputs,
each building on the previous one, leading to a more complex and coherent final output.
And finally, a difference in output. For prompt chunking, a single and more focused output is
generated based on the combined result of chunked prompts. However, for prompt chaining,
a series of interrelated outputs that form a cohesive narration or solution is generated.
While all these are being done, it is important to specify the output structure.
By providing a clear structure, you can ensure that the chatbot's response is well-organized
and easy to understand. If you want a structured response, tell the model how to format the answer.
For example, you can say, answer in bullet points, or you say, respond in a tabular format.
This is an example for a meta prompt for a marketing and communications executive for
campaign development. The last one we did was for strategy and planning
for a marketing and communications executive. So we are prompting chatGPT to say,
for, remember, we're talking about meta prompts. We're saying, act as a marketing and communications
executive at the hiring and talent management firm, and design a multi-channel campaign
to increase brand awareness among millennials for financial services company. Also, I want you to
develop a promotional plan that targets busy professionals. I want you to also conceptualize
a trade show booth experience to showcase the services of the hiring and talent management firm.
Let's see what the response is. As we can see, the response from the prompting gives us this
incredible answer. Number one, it says, a multi-channel campaign to increase brand
awareness among millennials for financial services. The objective, it shows that.
It also shows the campaign theme. It suggests a campaign theme. It tells about the key messaging
that can be used. It talks about channels and tactics that can also be used. So you see,
for social media marketing, the platforms to use, the content strategy that can be used,
the financial tips and hacks, the influencer partnerships, the content campaign as well,
Instagram stories and posts, and these ideas that the chatbots is generating based on the prompts
that are given. And remember, we are talking about meta prompts. You also see the content
marketing response that it gives that can be taken. These advices can be taken by the marketing
and communications executive to bring the objective to life. So based on the prompts,
it also shows the promotional plan for busy professionals. It shows the objective,
the target audience. In the target audience, you can see it's saying to, as a response,
that you can target IME professionals, entrepreneurs, individuals in demanding
industries. It gives an age group, it gives priorities as well, and these are professional
plans that can help busy professionals. It also gives promotional tactics that can be used
to reach these professionals. So some of the promotional tactics that the response gave
is personalized financial assessments, exclusive consultations, content partnership as well,
guest speaking, and these are ideas that the marketing and communications executive can use.
You can tweak, you can make better, you can take some things out, but these ideas that have been
generated by the chatbots for the marketing and communications executive. You also have ideas
around time-saving solutions, how to create a message that emphasizes time-saving benefits
of working in financial organizations or working with financial experts. You also have email
newsletter on how to reach a lot of these people in this segment. Then you have the third idea
based on the prompts that was given to conceptualize a trade show booth experience,
and you can see there it even gives a trade show theme, right? It gives you how to design the booth,
the interactive touch points, how you can make the live demo sessions upon a financial health
check station as well, right? Even the response is saying how you can set up kiosks with tablets
where attendees can take a quick financial health check, a quiz, and also after completing it,
they get personalized insights and digital reports summarizing their financial situation. So
all this can be used as a form of campaign development for the trade show booth experience.
And remember, we're still talking about meta prompts. Another idea the chatbots gives is a
virtual reality experience using VR headsets to immerse attendees in a simulated journey of
financial growth over time. Through VR, they truly can experience how financial planning is today.
You also see photo on social media wall where you can create a branded financial goals
where attendees can write down their financial goals. You can also create a branded financial
goals wall where attendees can write down their financial goals or dreams on large visually
compelling display. And this encourages them to share their photo on social media with specific
hashtags to enter a contest, to win free financial consultations, or whatever idea that the
organization is looking to bring to life. So you also see another idea around on-site financial
experts, giveaways and branded merchandise, and engaging in presentations and panels.
All these ideas are generated by the AI chatbots using the meta prompts
for the marketing and communications executive. So depending on the field you find yourself,
you can always prompt these chatbots to give you ideas to work on and this reduces your
thinking time, reduces the time you spend on ideating or even fleshing out your ideas.
And these chatbots have been programmed and processed to give you a response that is
befitting to what you require or desire. But this starts from the right prompting.
You can see the chatbots even giving campaign metrics and KPIs, how to track social media
engagements, how to track website traffic, conversion rates, boost interaction, even
customer acquisition, based on the prompts that I was giving. And this is an example
of meta prompts for the marketing and communication executive for campaign development.
Platforms like ChatGPT, Google Gemini, may feel private when you're chatting,
but they are not private in the way you might think. Your conversations can be stored,
reviewed, and even analyzed to help AI models learn and improve better. This means that whatever
you type into these tools should be treated as public information. So even if you're working
on something sensitive or brainstorming a confidential business strategy, do not assume
that your data is secure or encrypted. It is not end-to-end protected. Think of these platforms
like sending a message through a helpdesk system. Yes, it is smart. Yes, it is helpful.
But someone at the back end might see what you have typed.
Hello again, and welcome to Module 5, Security and Safety While Using LLMs,
Large Language Models, and AI Tools. In this module, you will learn about caution in relation
to LLMs and AI chatbots. Let's dive right in. While these AI tools can help us optimize processes,
write, summarize, plan, and even code, we must remember that they are still cloud-based platforms,
not private notebooks. So what are cloud-based platforms? Cloud-based platforms are platforms
that allow organizations to store, backup, and recover data. It is basically a large computer
with a huge storage owned by another big organization like Microsoft, what you have
with Azure, what you have with Google, which is GCP, which is known as Google Cloud Platform,
where many other companies store their information. Although the platforms guarantee safety,
it is important that we are knowledgeable in scenarios. Let's dive into key
things every user must know while using LLMs and AI tools.
Let's get through some examples of information you should never enter into an AI tool like
ChargeBT, Gemini Cloud, Loveable Copilot, VO by Vessel. Never share personal identifiable
information, PIIs, like your full name, your address, your phone number, your national
identity number, your banking details, your health records, and more information that can
be easily used to identify you. Also, never share client or company data, especially if
you work in industries like finance, law, healthcare, and tech, or you work in a government
institution. Also, never share internal or confidential documents, including strategies,
product plans, legal documents, or anything marked confidential. Login credentials, passwords,
or access tokens should also never be shared with AI ChargeBot. Let's say you're a lawyer
drafting a contract. It is fine to ask an AI to explain the differences between a will and a trust,
or a lease and a license, but it is not okay to paste your client's private documents or names.
As we say, if it's private offline, it should stay private online too, even with AI.
Now, let's talk about ethics and responsibility. Just because an AI can generate content doesn't
mean you should ask it to do anything and everything. Here are some examples of what
not to do. Do not use AI to generate harmful content like hate speech, threats, or harassment.
Do not use AI to spread false and misleading information. Do not try to impersonate someone
else, even if it's just for fun. Don't create deep fakes, fake testimonials, or fake news using
AI tools. Remember, every prompt you write reflects on you. Use AI with integrity, and remember,
you are responsible for the outputs you guide the models to produce.
An AI ChargeBot is simply a tool and doesn't have values, but you do.
Let's talk about some practical ways to use LLM safely, especially in a professional or work
setting. Use company approved AI tools. If your company has specific tools like Microsoft Copilot
or a custom GTP setup, use those instead. They may already have security features added to them
and internal data policies. Another best practice is to disable chat history when needed. Most
platforms allow you to turn off chat history and delete conversations. If you're working
on something sensitive, you can disable chat history before you begin. Review the terms of
services. Always understand what the platform says about data storage and usage. These things
change often, so make a habit to check. Another strategy to use is to use placeholders for private
data. For example, instead of writing James Bond's loan amount for 1 million naira, you can
say client X's loan amount. LLMs are helpful copilots but not vaults. Share only what you
would be comfortable saying in a meeting with strangers or in public transportation.
Even if a platform promises privacy, AI generated responses are currently not encrypted end-to-end
like secure messaging apps. This means anything you generate, including your prompts
and the AI response, could be seen, stored, or reviewed by others. So here's a rule of thumb.
If you wouldn't post it on a public blog, do not put it into an LLM. For example,
don't enter employee salary records, don't enter medical diagnosis or personal health data,
don't upload legal case files, or more sensitive information.
In this module, we've learned that LLMs like ChargePT, Gemini, are not private,
they are cloud-based services. You should never share sensitive information such as personal data,
company secrets, or passwords. Responsible AI means not generating content that's harmful,
misleading, or unethical. There are best practices you can adopt like disabling chat history and
using company-approved tools. AI responses should always be treated as public and not private.
Remember, AI is powerful, but with great power comes great responsibility. Whether you are using
ChargePT to draft an email or Gemini to brainstorm business ideas, enforce your ingenuity to think
before you type. Use your knowledge, use your judgment, and always use AI safely.
You are in charge and not the AI.
Introduction to Advanced Prompt Engineering. Welcome to the bonus module. I have created
this module for you to learn about an important Advanced Prompt Engineering technique. Let's get
right into it. Let me introduce you to a concept called multi-step prompts. This involves giving
the chatbot several instructions that build on one another. For example, you can first
ask for a summary on an article, then ask it to critique the summary, or you can ask it to first
write a story outline, then write a detailed plot, and finally write the full story. This method
helps to improve the coherence, relevance, and creativity of generated output.
Important things to note. Ensure a conversational flow. To create a natural and engaging conversation,
it's important to consider the flow of the dialogue. You can use techniques like back and
forth questioning, summarizing previous points, and introducing new topics to keep the conversation
flowing smoothly. Another thing to note. Properly handle nuances and idioms. Human language is full
of nuances and idioms that can be challenging for LLMs to understand. When working with these,
it's important to either define the expression or rephrase it for clarity. To improve their ability
to understand and handle these complexities, you should provide as many contexts as you can.
Using clear and concise language. For instance, instead of saying, it's raining cats and dogs,
you could ask, it's raining heavily. What are some of the idioms that describe
heavily? Another important thing to note is dealing with circling and humor. Remember, AI comes to be
based on the training of competent systems to understand data. This makes it difficult for
chatbots to understand circling as it relies on two contexts that are to grasp. To get better
outputs or responses in this regard, you should be explicit and provide clear examples. If you want
the chatbot to respond with humor and circling, you can say, respond with a sarcastic tone,
or I want you to be sarcastic. In this assessment, we are required to create a real-world
problem or challenge, such as education, poverty, or healthcare. Develop a series of prompts to
guide an AI model in generating creative solutions to this problem. Use the examples in the next video
for guidance. So let's take an assessment example. Let's say you are tasked with designing a multi-step
prompt to plan a new app launch for your tech company. The goal is to ensure that the launch
is well-coordinated, the team is prepared, and any other potential challenges are accounted for.
In the following steps, you can generate a credible plan via an AI chatbot. Step one is to generate
a task list. You can create an open-ended prompt that asks the AI chatbot to list the main steps
for launching a new app for a tech company. Step two, you can expand on the task. Use prompt
chaining to build on one of the tasks identified in step one, asking the AI chatbot to break it
down further. For example, ask the AI chatbot to explain how to create a marketing campaign
for your app launch. Step three, you can apply Qshort Learn. Provide the chatbot with two examples
of marketing strategies for other tech launches. You can use previous examples of strategies you've
worked on, then ask it to generate a similar strategy for the new app. Step four, use guardrails
and specify output structure. You can ask the AI chatbot to format the strategy as bullet points
and limit the response to one to two words, or however the number the length you prefer.
Step five, you should anticipate challenges. You can use a meta prompt to instruct the AI chatbot
to think like a project manager and predict two major challenges that may arise during the launch
along with the solutions. We have now come to the end of the course. I have taken you from the basics
of artificial intelligence into prompt engineering. By now, you should have a solid understanding
in creating effective prompts and understand how to get the best results from AI chatbots.
Keep experimenting with different prompts and soon,
you'll master the art of prompt engineering. Remember, practice makes perfect.