Transcripts
1. Introduction: So Deepsek has taken the artificial intelligence
industry by the storm. In this current class, we
are going to be diving into the application and
the buildup of Deep Seek, understanding its
core models and functionalities in which you are going to be equipped
with exclusive, up to the market,
up to the trend, insights and information
and knowledge base to help you get up or running with Deep Seek with
fraction of the time, getting you equipped with essential knowledge in terms
of how to use Deep Seek for various applications and
the key differences in its functioning as
a language model, as an artificial
intelligence model compared to other artificial
intelligence models out that way, you have a
clear idea how to use it, when to use it, and
its court strengths, and the areas that
it will help you excel in specifically
within the current trend, within the current industry of artificial intelligence,
and much more. All of these crucial milestones, we're going to be covering
them in this current class.
2. Your Project: Your project for the class
revolves around diving into Deep Seek and finding out the key differences between
its various models, V three, R one,
search capabilities, and combining all of them in order to refine your output in which you are going
to be creating various outputs for
comparison purposes, using its models based
on a single prompt. You're going to take the
same prompt, build it, craft it in the best
way possible using the prompt engineering practices shared with you in
this current class, and you are going
to fine tune it through the integration
of the various models, starting with V three, going into R one, combining
it with the then you're going to be
sharing your results with the rest of the
community for feedback.
3. Setting Up Deepseek: Now, in order to get up and
running with Deep Seek, we need to access Deepsek. The first thing that
you need to do is go to the website which
is depseek.com. Once you type in deeps.com, this is the main homepage, and you're going to have
two main options over here. You got the Start now and
you get the deepsk app. If you're using
your mobile phone, once you click on this, you have the ability to download
the deepsk application. And once you click on Start now, you have the ability to just simply jump into the interface. Using desktop, you'll notice that you have no ability
to click on this, but you're able to click
on the start now option. You have the ability to
scan the QR code directly from your desktop to download
it on your mobile phone. But as you go about
beginning with Deep Sk, you're going to click
on the Start button, and then you have a
Deep Seek V three, which is the basic version
or model for Deep Sk. But before we get to it,
let's explore the main page. As you can see, Deep
Seek has a clear table reflecting the capabilities
of deepsk the models compared to other known
models such as Cloud and GPT based on different
criteria English, coding, mathematics, Chinese, since Deepsk is
developed in China. So as you can tell, these are the basic comparisons
across various parameters. Obviously, we're not going
to get into that because it doesn't serve us any
purpose at this current stage. So what we do right now, once
we get to the landing page, we click on Start now. Once you dive in to deepsk
and you click on Start now, you will be prompt
to actually go ahead and sign up for Deepsk
keeping in mind, since Deepsk has been
getting a lot of attention, you will notice that
sometimes the server will lag or it will show you
that the server is busy. You do have the option
either to go with email address or a
login with Google. Here you got the option
for phone number as well, and then you are good to go. So go ahead, take a moment
to set up your account, and we are going to proceed
up next with utilizing Deeps.
4. Exploring the Interface: Welcome back. So this is the first prompt
window for Deep Sk. It looks quite
similar to Chachi PT. I'm going to explore
it before we actually get to the
applications part. In addition to learning about the various models
that you're going to be dealing with when it
comes to using Deep Seek. So by default, when you
are landing on this page, you got I am Deep Seek, and this is the prompt window. And you will notice by default, this is the Vo three or the V three option
from Deep Seek, which is the basic model to be used. Here you
got two options. You got the Deep think or
one and the search option, we're going to be
exploring every single one of them with details. But Depthk R one is the one which is
getting a lot of hype recently because it goes
into the expert mode, sort of speak to get
you powerful results. Now, let's explore
the interface. If you navigate
to the left part, you open the sidebar
and you have the ability to start a new chat, and then you have the ability to get the application
and you can download it to a mobile phone
once you click on New chat, you'll be prompted
back to this window, and you are able to actually start engaging with deep Seek. So we're going to start off with a very basic prompt structure, and I'm going to
show you some of the basic capabilities as we build up our level
with using deep seek. Keeping in mind, I have the ability to toggle
these options. Notice once I hover over
Deep Seek or One, it says, use Deep seek or O to solve reasoning problems because this involves complex thinking. Over here, search the web
when necessary very powerful. Why? Because you have
the ability to get live information straight from the web as if you're
using Google for search. Now, if you've
been using hatGPT, you would notice that Chat
GPT actually is trained. Even on the search model, it's part of an add on. You have to subscribe
to the paid version to be able to search the data, and the search data is not
live, it's not up to date. There's a certain cutoff period. So that being said,
Deep C gives you the ability to search for data live information, website links. At the same time,
conduct deep thinking, complex calculations
as part of reasoning. It's part of the
buildup of deepsk. For HAGPT, for example, even though you got
GPT four oh and three and whatever model
that you have for CHAT GPT, but the buildup or the
architecture for JAGGPT is made up of different
allocation for the resources. And the search feature is
not innate. It's an add on. It's something that you
add to the architecture. But as part of deepsk,
this is part of it. So if you search for something, you have the ability to
actually retrieve it. So I'm going to be
showing you through the various usage for these
models one step at a time.
5. Deepseek Downtime: Back. Now, this is an important lesson
that we need to keep in mind. Since Deep Seek is
quite free to use, you might experience
some sort of service disruptions
in case the servers are the servers are
not able to handle it. So how do we go about
identifying this in terms of what are the best timings
to use Deep Seek as of now? Obviously, this might be
solved in the future, but we need to pace our timings to know when
could we use it and when it's facing problems and
downtime to save us the trouble. So we go to the
homepage, deeps.com and scroll all the way to the
bottom. Take a look over here. Now, you got the
various descriptions for the various models. You can explore
that at your own. As for the products, you got Deepsk applications,
Deep Seek chat, which is the one
that we have right now, Deep Seek platform, the APIs, and the
servers status. You click on the server status. This will take you to
the following page. You could subscribe
to the updates. You'll get notified
on your email with all of the updates, but notice what's
going on over here. Deep Seek performance
is being affected since there are a couple issues taking place with
lots of traffic, scroll all the way down.
Notice what you got over here. Take a look at this.
Now the performance as of now, this is the red area, it shows you that there
is some sort of downtime, major outage for how long partial outages
has been running perfectly. Everything is perfectly
fine. But this is expected because the Deep seek artificial
intelligence model has been launched and it's taken the artificial
intelligence industry by the storm within the past
couple of days and weeks. So it's an influx of
users to the servers. That's why the current
outage is justified. So the green bars reflect that the servers
are up or running, but here there's an outage. And the major outages
has taken place for over 24 hours so far. So if you've been using
Deepik and you tend to find the description to show you that servers are busy,
try again later. Take a look at the
service status. That way, you have a clear
idea if they are facing any technical issues or
there's any downtime.
6. Deepseek V3 Vs Chatgpt: Seen a basic overview
of Deep Sk interface, and the basic model
is deep Sk V three. This is the basic model, we just simply when you
land on Deep Seek, they're going to be using
as part of the interface. So in this current
lesson, we're going to understand its features, the Deep Sk V three model
compared to CHAD GPT, as well. That way you have
a clear idea when to use every single model. So over here, we got a table. We get the features,
we go Deep Sk V three, and we got HATGPT. First of all, for
general knowledge, deepsk V three gives
you up to date information such as
current events trends. It has live search
feature capabilities. We've seen on the
interface, you have the ability to click on search, and literally like Google is able to fetch
live information. However, for CHAD GPT, let's say for the free version, you don't have the
ability to do so. And for the paid version,
the subscription, you have the ability to search, but there's a cutoff
period, usually, which is September
2023 could change, but there's a certain
time frame because as part of the
architecture for hATGPT, it doesn't have the search capabilities
compared to Deepsk. As for coding and programming, Deep Seek V three specialized
in Chinese tech ecosystems, for example, Anibaba, Cloud, wich hat, and API. Since it has been
developed in China, it knows all of the coding
used within the region. It's familiar with that. However, for Chad GPT, it has a strong general
programming approach to Python Javascript
and open source tools. But it's important to note that even Deepsek has the ability to take into account
Python JavaScript, even better than had GPT. But the thing is, we're focusing on the key
distinguishing factors. As for creativity, it's optimized for Chinese
cultural context, if you're trying to come up
with certain themes, songs, idioms or trends, it will draw some inspiration
from the context it has been developed in. Cha Ji PT, on the other hand, since it has been developed
on the Western culture, you got the western centric
stories and analogies. Based on the
geographic location, every single one
of them has been developed in a certain
region, right? So the model has been trained by individuals living
in those regions. That's why you tend to find
that the creativity or the cultural impact is fused
in the training process. As for the math and logic, Deeps is very powerful. It's competitive and
problem solving with focused on practical
applications, especially when you click
on the R one model, like you put it on overdrive. It goes for the practical
applications and the analysis like
little calculations. I will show you how you go about the thinking
process that we're going to see once we dive into
using the R one model. As for a Chad GPT, it's strong in the
theoretical part and step by step applications. But the actual calculations, it falls behind slightly a bit. I've tried ChanPT for
mathematical calculations, I will show you the sequence, the logical steps that
you need to follow to solve the problem like step one, do the step two, subtract that. But once it goes about the
calculations, it falls behind. Literally, it could conduct
false calculations, wrong conversions, and
give you wrong results. But you find the deep
seek to be stronger than that as for
language support, Deep Seek prioritizes
Chinese fluency, localization in the
form of the slang, the writing terms used, it's drawn from that background. As for haGEPT, it has more
multilingual support, but less nuanced for the Chinese language since both of them are developed in different parts of the world. So this means the
training has been conducted with
different approach when it comes to languages. But both of the Beep
Seek and ChaGPT, they are powerful in terms
of language support. But like I mentioned, these are small tweaks differentiators to keep in mind as you go about either using Deepsek or ha GPT. As for the user experience, the process is streamlined for technical and business queries such as data analysis and APIs. Since Deep Seek focuses on the technical
applications, coding, mathematical calculations, you tend to find
that businesses lean heavily into it and technical
queries, such as coding. However, Chad GPT
is more generic, it's more conversational with focus on general
purpose dialogue, list, create, respond to an
email, write an article. These are some basic
applications that you don't need sophisticated or in
depth technical knowledge. As for the integration, Deepsk has been designed for
the enterprise use cases, literally businesses they need the power of technical analysis, such as cloud services, deboPs and workflows,
Deepsk V three, the free model can deliver
those as for Chat GBT, it's best for a personal
educational use and third party
app integrations. Something basic doesn't need so much technical
in depth analysis. You can use Chat
GPT for it based on the applications at hand. If you're going for
something mathematical, engineering base, and
whatnot, go for Deep Seek. If you're going for something
which is conversational, summarizing, finding
errors, grammars, comparing, for example, responding to certain
features or emails, whatever it is, you
can go for Chat GPT. Response style is concise, technical, and task
oriented, literally. Once you go about deep seek, as we're going to be saying
for various applications, the response is more of
straight to the point. Less fluff in terms of the language,
straight to the point. Bullet points, short paragraph, short summary gets you to the result in the less
number of steps as possible. However, for chat GPT, more verbals, explanatory
and narrative. You tend to find once
you go about Cha GPT, most problem that explains
and talks and tries to provide words to provide
some sort of a mood, a context rather than being quite too technical
or formal speak. You have the ability to tinker
with the tone of Cha GPT, which Deep seek lacks
behind in that case. These are some core differences in terms of deep Seek V three, the model, which is
the basic version, the basic model versus ChaGPT, the basic model as well, not the paid subscription. Once you go about dealing with deep Seek, you
have a clear idea. Are the areas of strengths? What are the areas of
weaknesses such that you can decide you go for deepsk for this application
or for CHAD GPT. But there are the other models
such as the R one model, which is the powerhouse of Deepsik which we are going
to be discussing as well.
7. Deepseek R1 Vs Chatgpt: Now the other model
that we have for deep seek is the
deep seek R one. The previous model, the
basic one is V three, and this one is R one,
which is the deep thinking. Model. Both of them
are free, by the way, but this one is more powerful and for specialized purposes. Let's take a look at it
as for deep Seek R one, we're going to follow
the same analysis compared to HAGPT. For the technical focus, deep Sk R one is optimized
for technical queries. Notice it's always
about technical parts when it comes to
deepsk such as coding, math, data analysis, and engineering where you
need calculations. ChaGPT on the other hand
is for general purpose, dialog and creativity, articles, reading, comparisons,
brainstorming. It all stems from the difference in their model architecture. Since deep Seek is MOE,
mixture of experts. Focuses on certain resources
to solve a certain problem. However, CHAT GPT is
dense. It's dense. It means it covers
all the resources for any analysis before
coming up with an output. So we tend to find that
deep seek works best for technical queries because
you need to focus on the experts sort of speak, expert resources for certain
engineering applications, mathematical applications. However, Chat GPT, since
it goes through all of its training resources
to come up with an answer, it's better suited for general purpose dialogue and
creativity applications. As for decoding support, you tend to find
that Deep Seek RO specializes in tech ecosystems, but it's still Python, Java, GML, it dominates on
that area as well. JAGGPT better for
global frameworks as Python JavaScripts,
and react. Both of them, we have
those capabilities. But you tend to find
that Deepsk has the ability to
debug codes easily, even write codes without
bugs in the first place. But if you try HADGPT for
writing a piece of code, times you notice that once
you try to run the code, you find errors and bugs. And if you try to
reiterate the process, it might be time consuming. So for coding, you tend
to find deep seek to have the positive or the
upper hand in that case. As for math logic,
for deep Seek, it prefers or
prioritizes step by step problem solving for stem topics like science, technology, engineering,
mathematics. As we see with deep Seek
are one, literally, it shows you the
thinking process of the model before
giving out the answer. As for Chat GPT, just simply gives
you the answer. But for deep seek R one, you get to see the
thinking process as you go through it and once you go through the thinking process, you're able to find errors in this and you're able to
reiterate the process. That's very powerful.
Now for Cha GPT, there's strength and
theoretical explanations, but less optimized for
niche technical tasks. This is very true
because hA GPT if you use it for any technical
aspects such as engineering, mathematics, sciences, it will give you the
explanations for the theory, right, steps,
sequences, analysis. But when it comes to
the pure calculations like lural calculations,
it falls behind. It makes mistakes. As for
the language fluency, you tend to find
both of them, they are fluent in languages. Some of them are
leaning, let's say, deep seek towards the cultural
context, generated from, and deep Seek also from
the cultural context, haGGPT from the
cultural context it was generated from
compared to Deep Seek. So the training has been
conducted by individuals. So obviously, there
is an influx of different backgrounds in the model as part
of the training. As for the response style, you tend to find a deep seek. It doesn't have some
sort of a mood or a certain tone to it. It's concise, structured
and task oriented. You cannot tell if it's
using a friendly tone, serious tone, professional tone, but with chat GPT, it's
more conversational. So you have the
ability, as you get to learn about prompt
engineering practices, you're able to change the tone. For example, write an email in a professional tone in a
friendly tone in a formal tone. Chat GPT has the upper
hand in that case. As for the up to date knowledge, this is a very important part. Deepsk R O has access
to fresher data. 2023, 2024, trends, tools, and APIs because
it has the ability to search live for
the information. However, for HA GPT, there's a cutoff period.
It's like a model. You give information,
you train it, you cut off the information. Then you give another
source of information, you train it again, then you
cut off the information. But for Deep Seek, it's live. Just like searching on Google, you tend to find
something by typing, Deepsk can do the same thing. As of the usage cases
for enterprises, you tend to find Deep Seek for business and
technical workflows. However, HAG GPT
is for personal, educational use and
creative projects. This explains the difference
because Deep Seek is built on a mixture
of experts model, focused resources
for focused tasks. CHAD GPT, all resources, depending on the
prompt, are going to be used to help you
with your task. As for creativity,
Deep Seek one is focused on practical
outputs such as reports, coding, calculation,
sequence of steps, analysis, data analysis,
more of a technical part. However, for chat GPT, it excels in
storytelling analogies and open ended
ideation because it has all the resources being used for that
specific problem. So it doesn't allocate
certain context to it. It uses all the resources with all contexts to get information to help you
with your application. So that's why we tend
to find that it's more open ended with the
ideas and their creativity and the options and the
conversational style with chat GPT as you go
about typing it. However, with deep Seek, since once you give it a prompt, it focuses on a resource
or a couple of resources, which are called as
experts to help you answer that question in
the best way possible. So deep Seek R one is considered to be more the
deep thinking analysis. Of deep seek to help you conduct sophisticated
technical analysis. Think about mathematics,
calculations, data reports, data analysis, engineering practices,
practical calculations, financial calculations,
all of these things which require
technical expertise. Deep Seek will help you
out because Chad GPT will show you how to do them
like the steps in theory, but to conduct a calculation where one plus one equals to, for example, you'll tend
to find HAGPT will have an issue if the calculations
get quite too complicated, where you do have a
lot of conversions, a lot of steps in the process. Like we have mentioned,
Deep Seek has a bigger one, two, eight K tokens, conversational window,
it has the ability to retain more information as you go through the
conversation with it, which helps it in the
technical calculations part. However, hATGPT, the
limit is smaller. So let's say it slips on the
information along the way. That's why it affects its
calculation capabilities.
8. Powerful Deepseek Search Capabilties: Important to highlight
the search capabilities, this topic is often
overlooked when comparing deepsk and CHAT GPT. When you are using
Google, for example, you go in and top a
keyword and you're able to find certain links
and websites, right? If you go on HAGPT, the free version,
you cannot do this. For the paid version,
GPT four, for example, the add on as part
of the subscription, you have the ability to search, but it's not part of the buildup of the
architecture of hATGPT, so the information
that you're getting is not actual live data. However, with Deepseek the story is different.
Let's see how. In terms of the search
capabilities for DeepsekRO, accessing the web is real time. Once you go on the
search, click on it. As part of the prompt
window, click on search. It means go ahead and get
information live right now. However, with change GPT, for the free version,
you don't get it. For GPT four, there's a
combination with Bench, and you're able to get
information, but it's unstable. Sometimes it works, sometimes it doesn't it has been discontinued
for a period of time. But with GPT four, the
subscription model, you're able to actually search, but the information
is not reason. The website could be down, but the information
is still out of date. Deep Seek will get
you the live data. Think about it like you're using a search engine, the same way. For Chat GPT, the
search has been trained to give you the data it has stored over a period
of time, so to speak. Data freshness for deep Seek, you can retrieve live
information, for example, 2024. But for chat GPT, limited to pre 2022, data less using Ben
for live updates, like I've mentioned, has been discontinued for
a period of time. It's quite unstable
in that case. So GPT, hat GPT. Even for search features, it doesn't get you
real live data. It's also trained because it's like an add on,
it's architecture. However, with deep seek, it's within the architecture. It has the ability
to fetch live data. As for technical queries, deep seek is
optimized for coding, math, and live data. This is the core functionality of Deep sik in
terms of using it. If you're planning
to go for Deep sik, you got something related
to coding, mathematics, anything with calculations, searching for live data,
deepsk will help you out. As for creative elements such
as storytelling, writing, marketing campaigns,
ideation analysis, which is quite generic
and theoretical, you can go for the hATGPT
application, keeping in mind, you will not have the
same search capabilities for live data
compared to Deep sk. But obviously, hATGPT still
can get you information, but the information
freshness is not as fresh. Now, this could be
changed in the future. This could be modified. Still, the whole thing
is kind of live. And we are sharing the
latest information, the latest trends in terms of artificial intelligence for
Deep Seek and CHAD GPT, as of now, to help you
get up to speed and up or running with the
market trends and shifts. That way, you're
getting equipped with the expertise and skills to help you stand out in such a market with
fraction of the time. As you're watching this course,
this is the first course actually to go about the
entire subject with details. That way, you're saving
time, saving effort, trying to acquire
such expertise in a rapidly evolving
industry and market. So at this current
stage, you have a clear idea about
the strengths, weaknesses of Deep Seek and HAGPT and how they
compare with each other and the various models for
Deep Seek that you could use, as well as their own
limitations and strengths. That way, when you are trying
to use any of these models, deepsk HAD GPT, whatever it
is, you have a clear idea. Take a look at your application.
What do you need to do? And based what you
have learned so far, you have a clear idea which
one that you need to use, whether deepsk or HAGPT.
9. Prompt Engineering Practices Part-1: Now we're going
to be diving into the prompt engineering practices when you are engaging
with Deep Seek. Since it's based on the NLP format for engaging with the artificial
intelligence model, we need to use prompts, and there are certain ways to go about creating the prompts. First of all, when you are
dealing with deep Seek, we do have a series of
prompt Jing best practices with some examples to help you understand how
to interact with it. Even if you're a complete
beginner, this will guide you. First of all, be as clear
and specific as possible. Why? Because vague prompts
they lead to vague answers. How does that look
instead of saying, tell me about AI,
this is the prompt. Instead of putting this
prompt on Deep Seek, go ahead and use
a better prompt, which is explain how AI is used in healthcare
diagnostics with examples. Always be specific
and be action based. Explain, write, elaborate. This is part of being
clear and specific. Provide context. This is very important. Why? Context helps the AI tailor its response
to your needs. And what does that look like? Instead of saying, what is
the best marketing strategy? Logically, if you're
speaking to someone, you tell them what is the
best marketing strategy? You need to provide more
contact, more information. So a better alternative, a better prompt would be I run a small ecommerce store
selling handmade jewelry. This is the context, the
background information. What's the best
marketing strategy to reach millennials
on Instagram? Notice how specific it is. I started off by giving a context to explain
the background story. Then you provide
the instruction. What would you like it to do? That way you get finer results
with minimal iterations. Now, we do have a sequence of prompt engineering
practices that we need to get out of the way, and these are the first two.
10. Prompt Engineering Practices Part-2: Continuing to step by
step instructions. This is very powerful
and often overlooked. When you're creating a
prong, first of all, make sure that you break it
down two steps to ensure a thorough response as you
are trying to think out loud, input the prompt
as the process of thinking out loud to help
you get better results. So how does that look
like? Instead of saying, how do I start a business? This is very vague,
very generic. Why not go for,
provide a step by step guide to starting
restaurant business. That's one part. That's the first
step, second step, including legal
requirements, that's two. Tools, three, and marketing
tips, that's four. So as you layer your steps, get me step number one,
then step number two, then three and four,
in one prompt, you have the ability to actually get the answer from one shot. Then you got the
specifications for the format. Why? The AI can tailor its
output to preferred structure. So what kind of format would
you like the answer to be? Detailed explanation,
brief summary, bullet points,
that's the format. So instead of saying, explain photosynthesis, very generic. Why not go for explain
photosensesis in three bullet points with simple language
for a 10-year-old. So here you've
gotten some context, you've gotten the requirement, the action based, and
that's the format. The bullet points. You could go for a
summary paragraph, three lines or a full article. So this is the format that you would like
the answer to be, which is part of prompt
engineering practices. Now let's transition to other prompt
engineering practices that you need to
be familiar with.
11. Prompt Engineering Practices Part-3: Continuing two of the most
important prompting practices that you need to be
equipped with constraints, which is often overlooked. Set limits set boundaries
for the model to help you engage at a level that you would like the output to cover, crossing a certain point, certain word limit,
certain number of lines and such forth. So constraints, they help focus the response and avoid
irrelevant information. For example, instead of saying, write a story, this is a prompt. Obviously, there are no limits, how many pages how many words, how many lines, how
many paragraphs. These are constraints which should be part of your problem. So use a 200 word story about a robot learning to
paint with a happy ending. So I've given it the context, context context plus
the constraint. The constraint is the boundary. So you give it the context
what you're trying to do, what you're trying to achieve. And then you put a framework
which is the boundary. Do not cross a certain line, do not cross a certain
number of words, a certain paragraph, a certain
page limit, and whatnot. Even the tone, maintain
a professional tone, avoid a formal tone, and all of these are
considered to be constraints. Also, this is a bit of a
prompt engineering practice. Ask for multiple options. Once you go for a prompt, in order to save you time
and effort of trying to reiterate and get
the same response over and over again
in a feedback loop, why not ask for multiple
options from the first go? For example, this will help us get diverse ideas or
solutions to choose from. For example, instead
of saying suggest a business. One idea. Why not? Suggest three low cost, fast to lunch business ideas in the education
sector, for example, or in the hospitality sector, or in the restaurant, business sector, whatever it is. So that way, you
have a clear idea. What are you trying to achieve? What are you trying to actually
get out of the problem? One option, two options,
three options, four options. I'll give another example. Suggest four ecofuel
friendly cars, for example. That's what I'm saying, suggest
an ecofuel friendly car, and then give another option. Give me another option and you repeat that prompt
over and over again, you can get this done with a single prompt from the get go. These are some solid
engineering practices. So far, we have covered a series of them up to this
current stage, and we got more
to cover up next.
12. Prompt Engineering Practices Part-4: Now there are additional key prompt engineering
practices that we need to be equipped with
in order to make sure that we're getting
the best out of deeps. First of all, we
need to make sure that we're using examples. Sometimes when you are
trying to use a prompt, often you dive into
typing, right? But you could have some
pieces of information, a soft copy of an article 0R a screenshot or basically
a table that you upload or a reference or a certain example of a tone that you could
add to the prompt. How does that look like? Of all, why do we
need to do this? Examples help the AI understand your expectations to save
the number of iterations. Instead of saying, write
a professional email, you're going to say, write
a professional email, that's the action to a client explaining a
two week project delay, which is the context. Example, tone, this is
the example of the tone that you're trying to use,
polite and apologetic. So this way, you're
able to showcase the narration to the deep seek artificial intelligence
model instead of trying to reiterate again and again based on the output
from the Get go. Give it examples,
give it the action, give it the context on the spot. And then once you
do so, obviously, you might need to still
iterate and refine. This is part of the
process considered to be a feedback loop.
Why do we do this? Because the first response
might not be the perfect one, especially if you are
training the deep Sik model for the first time or
your own application. So you need to
refine your prompt for better results,
and how do you do so? So if the response is too
broad, add constraints. Focus on the technical aspects only. This is a constraint. So instead of keeping a generic, when you get the
feedback from Deep Seek, add a constraint,
certain word limit, certain tone, certain format, in order to get better results. If it's too technical,
ask for simplification. Explain this in layman's terms. So you could actually change the output that
you're getting from Deep Seek by adding an iteration and refining
simply by telling it, minimize the technical
output or summarize it or simplify it or focus on certain element and
ignore another element. In addition to using
examples, and by the way, you could layer the prompt
engineering practices where you use examples, use context, use constraints, iterate and refine all of
them within a single prompt. But these are the
key layers as part of proper prompt
engineering practices, and we are not done yet. We got a couple important ones that you need to
be familiar with. That way, when you dive into the application
for Deep Seek, you have a clear idea
how to construct powerful prompts to get
the best powerful output, harnessing the technical
power of Deep Seek.
13. Prompt Engineering Practices Part-5: Now moving on to some of the most powerful prompt
engineering practices, and one of them, which is the role playing, leveraging the role
playing by itself, if you use such a prompt tactic, you will get really
powerful results. Why? Assigning a role helps the AI tailor its
tone and expertise, and you could try this
with deep seek since the whole architecture is made up on the mixture of experts. Once you give a
certain role to play, by default, you will minimize the effort it takes to channel
to that specific resource. This is advanced, and
this is still new and we are quite
ahead in terms of teaching you such
powerful tactics to leverage the technical
power of deep seeek. So how do we go about this? Instead of saying
explain block chain, we're going to say, you
are a tech journalist. This is what we call as
priming role priming, making it think as if
it's a certain role. A certain expert, by default, Deep Seek will pick up that specific
resource saving time in the iteration process. Explain block chain to a non technical audience
in three paragraphs. We've given it the action.
What do we need to do? The context and the constraint
in addition to the role. Notice how many
layers as part of the prompt engineering
practices we included in a single prompt.
That's very powerful. You could test it at experiment
as part of the process as well by changing the roles or by changing the constraints. Different phrasings can
yield different results, simply changing from
three paragraphs simply changing from three
paragraphs to nothing changes the whole output completely or changing
the specialization. You are a tech journalist,
change it to something else. You get a different output. So that way, you are
trying variations. And by trying variations, you could have different outputs based on your own preferences. For example, what are the
benefits of AI and education? How can AI improve learning
outcomes in schools? These are different questions
that you are asking. These are different iterations. These are different applications
within the same context. And that way, when
you get results, you're able to
find your results. Better and better as you are
engaging with deep Seek. That way, when you're
adding layers and layers to your own
communication with Deep Seek, using prompt
engineering practices, you have the ability to actually
get the best outputs and harness the power that
you have with deep seek, the technical power that it has. Now let's move on to other important prompt
engineering practices. Use action oriented language. Explain, do, sort, plan, create, and these are
the verbs that you use. Direct prompts lead to
actionable responses. Instead of what is
content marketing, create a five step
content marketing plan for a new fitness
application. Very specific. You got the context,
you got the action based verb at the beginning
of the prompt to do so. And then you do have
combination of tasks. This is very powerful and
this is often overlook. This will save you time by
bundling related requests. Instead of
communicating again and again and again with the
artificial intelligence model, you can bundle all of the
tasks in a single prop. So how do you go about this? Instead of saying,
write a blog outline, now write the introduction
very fragmented. You could say, write a blog
outline on AI in Education. That's one. Including
100 word introduction, that's too. So these
are two tasks. One for the blog and one for the introduction.
You can add more. Include a 200 word conclusion, include a copy of an article
that will be posted or published afterwards as much as you'd like to add
in terms of the tasks. Simply add a comma,
add the task, add a comma and add the
task, and that way, you're able to create one
solid prompt from the get go. Well, we're not done yet. We got more important practices in terms of prompt engineering that you need to be familiar with. Ask for resources and citations.
This is very important. As you are going about
technical prompts, you need to ensure accuracy
and credibility because at the beginning
of the training of the model, mistakes do happen. So in order to just simply
avoid using information as is, always ask for citations. For example, instead of saying,
explain climate change, just simply say, explain climate change and cite
three reliable resources. Specifically when
you're using deepsk, it has live search capabilities. So if it's able to
find the information, pick it up from the
web on the spot, this increases the
credibility of the information since it's
quite fresh and recent. Then we have what we call as the positive and negative
examples where you direct the response of
Deepseek by saying, Do like this, don't
do like that. So you give an example to
follow or an example to avoid. So this will help you
clarify what you want or don't want to do. And
how do you implement this? Instead of saying write
a product description, you're going to say, write a product description
for smartwatch. Example of what I
like sleek design, seven day battery life, avoid overly technical jargon. So you are directing
it by simply including the word follow
example of what you like, or example of what you should follow or follow
this as an example, avoid this as an example. So that way you're putting
deep seek in its lane. Adding a prompt, give it an
example, give it constraints, layering with a certain role, and then you add some
sort of context, and that way you
add an example and something to avoid and
something to follow. That way, you get full proof prompt engineering
practices to get the best results even from the first go, and
we're not done yet. We've got some
important practices that you need to be
following as well. One of the most important
ones is actually not to keep the loop open. As part of prompt engineering, it's a feedback process, so close the loop with
a call to action, direct the AI to conclude effectively what it should
do at the end of the prompt. Instead of saying
explain time management, which is vague and generic. Explain time management
techniques and end with three actionable tips
for students, for example. These are quite straightforward. It shows the end result
for the AI model. That way, it has a clear
idea how to navigate to it. And finally, this is where
you harness the power of deep Seek through using or leveraging
technical expertise. Deep Seek Excels
coding. That's one. Math. And technical tasks. So use precise technical
language for these queries. If you're using a certain code, use the language of the code, Python, HTML, mention
it in the prompt. If you're using mathematics, mention the nature,
algebra, calculus. If you're doing something
related to engineering, what kind of engineering
included in the prompt? Because this is the
powerful part of deep seek and you don't want to fall behind as you're getting
the best out of it. And then you have
localized contexts. So if you query evolve Chinese
tech or culture, specify. Explain we chat mini
programs for beginners. Now, since Deep Seek is
actually developed in China, you have the ability
to actually use localized context with
a couple of clicks, and this applies to
other regions as well in which you use
familiar applications, popular applications, popular
tools as a reference. For example, create a marketing
campaign for Instagram. Or we chat or Facebook,
whatever it is. And Deepsek has the ability to navigate the web and get some
important information about the popular platform based on
the context that you're at. So at this current stage, you have been equipped with about 15 plus prompt
engineering practices that you could use collectively
or separately in order to get the best
output from Deepsk.
14. Deepseek V3 Demonstration: Come back. Now we're
going to get to the practical part of
learning about Deep Seek, where we're going to actually
tinker with Deep Seek and explore the various
options that we could have. So once we navigate to
Deep Seek right now, this is the prompt window
that we have in front of us. We're going to start with a
very basic prompt by saying, write a blog post about the
importance of Wellness. Now, I'm using at this
stage the V three model. This is the basic
model of deep Seek. Let's click Enter. Now, keeping in mind,
make sure that when you're using Deep Seek at
this current stage, actually, to use timings when
there's no peak loads to avoid any discontinuity
of the service. Take a look at the response that you're getting from deep Seek. As you could tell, it's
directly getting to the point by creating
the blog post, where it's breaking
it down actually into the structure of a
typical block post straight from the first go. Addition to sharing
some important details towards the end of it as
part of the prompt response, the first thing that
comes to attention is how extensive is the response that you would notice
it includes a lot of details like the window that you have in terms of the tokens. Take a look at the amount of explanation that you're
getting based on the output, including furthermore
details in addition to insights drawn by the
artificial intelligence directly where you go through the blog post and take a look
at this segment over here. It closes out the blog post, and it actually includes key elements as a wrap up on the block post
from the get go, simply by telling it, I would like to go with the blog post. So it follows the natural
rhythm of things. Now, if you explore
this final segment, what does wellness mean to you? Share your thoughts and
tips in the comments below? We'd love to hear how you're embracing wellness in your life. Notice the segment over here. Why did this come up?
Because from the get go, I gave it a very basic
prompt to write a blog post. So since it's action oriented
and it's very technical, it develops the
strategy to go from the beginning of a prompt to the final result in
the best way possible. Saving you the
whole iteration and back and forth communication,
keeping in mind, I did not apply any of the best practices of
prompt engineering. Now, if you could
experiment with this and you could
incorporate, let's say, personalization where you act
as a writer, for example, you change the prompt, and
you can just simply mention that act as a professional
writer and write a blog, for example, you know that the result is
completely different. Now, what I would like to show
you in this current lesson is the layers of
the application. So you start off with
the V three option, which is the basic option, which is perfectly fine, right? Now, as you transition to
further complicated tasks, we need to learn how to go about the Depth and the
search variations. And this is where we are
going to build up on things. So the first level is
the V three option where you go about
creating a prompt. It could be anything.
We're going to see the various applications
for Deep Seek, where you could use it in mathematics, engineering,
and whatnot. But for demonstration purposes, this is the starting point. Feel free to actually
go ahead on Deep Seek and try to include some
prompt from your own site, try to experiment with
it and go through the response to avoid wasting time reading the
entire blog post. You can do this by
yourself, and you could actually experiment with
other applications. Now, starting off with V three, we're going to take it one level further where
we're going to incorporate the deep
thinking procedure, the R one, which
is coming up next.
15. Deepseek R1 Demonstration: And welcome back. In
the previous lesson, we just simply went
through Deep Seek, starting off with the V three
model and we have started the prompt interface with a basic prompt to write a blog post about the
importance of wellness. Now, look what I'm
going to be doing. If you click on Deep
think over here, now we're going to take it into a technical aspect of deep Seek. And what I'm going to
be doing right now, I'm going to level it
up even more by adding an additional prompt
and to show you the key differences in
terms of V three and Depth. So this is the prompt that we
have had right now, right? Write a blog post about the
importance of wellness. I'm going to copy this again for you to see the difference, and I'm just simply going
to be pasting it right now. Here we go. Now, notice the difference they're going to be getting in the results. Exactly the same prompt, but simply activated Depth. Here we go. If you click
Enter, notice what's going on. You got thinking, and this is the powerful part
about deep seek. It shows you how it's
actually going through the thought process of
replying to your prompt. Take a look at this. Okay?
The user asked me to write a blog post about the
importance of wellness. Again, wait, didn't you
just ask that before? Let me check the history. Looking back, the
previous interaction was indeed the same request. The user wanted a blog post
on wellness and I provided a detailed with sections
like pillars of wellness, importance and tips, now
they're asking again, why? See how detailed the
thought processes. Maybe they want a different
angle or perhaps more depth, or maybe it's a different
user with the same query, but considering the context, it's probably the same user. This is very, very powerful. So once you activate Depth, you're able to actually
see the back end process, let's call it the
thinking process of the artificial
intelligence model, as well as how long it took. See how fast it is 22
seconds it was able to actually go ahead and
comprehend the query, even though it's quite
vague and broad. And analyze and come up with a response to what
I've asked it to do. So if you take a look at the
response that has shared, feel free to just simply pause the lesson and go through
it with detail just simply to understand the process
of deep thinking one, how it's going at the back end, behind the scenes before
giving you the end result. Now, if you take a
look at the response, the first part and
the second part, you will notice there will
be some slight differences, if you take a look
at, let's say, the multidimensional
impact of wellness, scroll to the top
it's not here, right? It just simply starts with
the importance of wellness, then what is wellness,
the pillars of wellness. If you scroll down even more, you'll find out that
we have wellness really means the multidimensional
impact of wellness. Notice that the response
is more detailed. It's more sophisticated. It includes more details, as well as more structure, more content to it, as well as some practical steps to prioritize our wellness. And further, let's say we find result.
From the same prompt. This is the V three version, and this is using the R one
version, the deep think. And this is an application
related to creativity. The true power shows
when you are going through technical
calculations as we are going to see in
the applications part, you will be surprised how
refined the answer is. So we have layered our response, starting off with V three. Then we went into deep thinking. Now let's take in
one level further.
16. Combining The AI Models and Search Capabilities: And welcome back. So what we're going to
be doing right now, we're going to take it
one level further from the R one model in
which we are going to actually take the same prompt as is to show you the difference
in terms of its capability. So I'm going to copy and paste
the same prompt over here. Now, the difference would be, I'm going to engage
now Deep think RO, at the same time, the search feature. So what does that mean? So when I click
on Deep think RO, it's going to add extra level of technical
analysis. That's one. And when I click on the search, it's going to search
the web when it's necessary to find some different
pieces of information. So I'm going to be using
exactly the same prompt as there is no
difference whatsoever, but for you to understand
how could you utilize it for your own application to differentiate the output.
So let's click Generate. So once we click on Generate, it's going to go through
the thinking process, keeping in mind that sometimes the server gets quite
loaded with the prompts. So if you find the
server is down, it means it has been
getting a lot of traffic. But as of now, the service is up and running and you're able to get the response on the spot. So first of all, it's going
to engage the R one model, and it's going to show
you the thinking process as you are going to be so, okay, I need to
write a blog post about the importance
of wellness. Let's start by understanding
what wellness really means. So you can go through
the thought pattern for deep seek as it
builds up the response. And once you spot
something which is wrong, for example, along the way, you can stop it and then you could fine tune
your prompt again, which is something
that you could do better compared to Chat GPT, where you just simply
input a prompt and you get the result then
you reiterate. So as we're going
through about this, what you would notice is that the thought process is quite extensive compared
to the previous one. So now it's including more of a technical in
depth analysis and giving you the
block post compared to the basic V three model. So if you take a
look at the previous response that we have gotten, you would notice that
the block post that we have obtained includes
some sort of details, but it doesn't dive
into further analysis. So once you engage
the deep think R one, you're going to
activate the process of the extra layer of analysis, which is very powerful for coding and mathematical
applications. Once we get to the
applications of deep seek, you will see how
powerful this is. Could apply to other
creative approaches as well if you'd like to get
more fine tuned results. And the search feature, this is where deep Seek
acts like a search engine. So as you take a look
at the final output, you notice there are
some key differences in terms of the
analysis part where the actual blog includes the information from
the previous output, but it will add up
some layers to it. What I mean by layers, it means going to add up some more deep thinking
analysis if you take a look at, for example, the myths, reality and the simple steps for your wellness
journey and the conclusion, it includes more
in depth analysis. And it can use the search feature because the search feature is going
to be live, by the way. So you're going
to take a look at the live web resources like a search engine to
help you find tune the output even further
when necessary. This is very powerful. So what I'm going to be doing right now, since I have the search
feature activated, I'm going to just simply add the actual links
into the blog post. Let's take a look at this. Include the latest
updates on the topic. Within the blog post
and provide the links. So what I'm going to
be doing right now, I'm directly targeting
the search feature where I'm going to be asking the Deep seek model to actually find the latest updates on
the topic from the web. So within Chat GPT, the information from the web is part of the
training of the model. It doesn't have
to be up to date. There's some sort
of a cutoff period. However, with Deep Seek, it will actually
transition to the web and start to fetch the latest
pieces of information. Now, if you would notice here, it says that the server
is busy to do so. Let's try one more time. Go ahead. Here we go.
Include the search. So what I'm going to
be doing is I'm going to reduce the load
on the server. This is something
that you could do as a pro tip in case you find
that the server is busy. Since it's a free tool
is going to be subjected to large demand until the servers are actually
increased in size. So I'm going to click on
stop the deep thinking part because I don't
need it right now. And let's go on the search. Hopefully, this will help us
actually get results quickly because we are reducing the demand on the
servers in that case. That way you are
trying to actually get sophisticated with
the model itself, because if you know
how to use it, you could tailor your selection using deep think,
search, V three. That's the whole purpose of teaching you these
important concepts. That way, you're
able to actually tweak them in the
best way possible. So if what we'd notice is once I stop the
deep thinking part, I reduced the requirement from
the model and just simply helped it focus on
the search features and look what it has got. 30 a results. So once
you click on this, you have the ability
to see all of the search results that are currently up to date
on the topic of selection. And notice the
dates, for example, 6 February 2025,
28th of Jan, 2025. This is very, very powerful
because you're able to use Deep Seek as a search engine
to fetch you information. And then you have the ability to inject the information
into your results. That's very powerful. Let's take a look at the block
post after fine tuning it with the
search analysis. Notice what's
happening over here. Latest wellness trends in 2025. So I have the ability right
now to update my result, incorporating the search
feature, and guess what? You have the ability
to find the links and the resources
within your result. See how powerful this is? So instead of creating,
let's say, a blog post, where the response
is just simply based on the trained
model on the contrary, now you have the ability to create a unique
piece of writing, which fuses the ability for Deep think or the V three
model as well using your own tone in combination with the latest search results, and not just that,
you're able to take your one level further by actually finding
these resources. If I click on this, I
have the ability to find the resource and the
date of the resource. So if you scroll all the
way down now you have actually modified your
for this current example, the block post where you've
actually gotten the results, the output based on
the search analysis, including the citations
at the same time, you are using the latest
up to date resources. Notice what I've said
there. Latest site updates. Look at the trend 2025.
So it's up to date. So it has the ability to actually be used as
a search engine, not just simply as an
artificial intelligence model that has been trained and
based on the training, it gives you results. You can just simply click
on Search and type in whatever prom that you'd like to fetch a piece
of information, and then it has the
ability to share with you the various resources on
that specific search term. So you understand the
difference right now. Now you have the
ability to utilize Deepsek as an artificial
intelligence model to get your result in combination with a
search engine model to get you the
latest updates from the web and you're able to combine them to
create masterpieces, blog posts, articles,
whatever it is using the latest
pieces of information.
17. Wrapping Up: So what do you think?
I truly hope that you found the class helpful
if it helped you get up and running with
Deep Seek and acquire the latest trendy knowledge
in terms of the world of artificial intelligence
and how Deep Seek has been shaking the industry
of artificial intelligence. You have acquired
knowledge which is quite new, quite specific, and you are ahead of the
curve if you have actually gotten something out of this current class by the
end of this current class. So I truly hope that you
found it quite helpful. I look forward to receiving your feedback on this current class, and make sure that
you follow my profile for the latest
releases and updates, and I'll see you
in the next class.
Engr. Hussein Attié, Entrepreneur I Engineer I Educator
