IA de Claude: el asistente de IA que vas a usar realmente

1. Welcome to Claude AI: The AI Assistant You’ll Actually Use: Hello, and welcome to the course on Claude AI. An AI assistant so intuitive, it feels like working with a seasoned professional who already knows your industry and needs inside out. Are you working professional looking for a smarter way to tackle your daily tasks? An entrepreneur balancing too many heads and in need of an efficient assistant or simply curious about exploring the potential of AI and wondering how it could help you work smarter, not harder? If you said yes to any of the above, I created this course for you. My name is Anna, and I'll be your instructor and mentor for the course. I'm line instructor with my other courses available here on the platform, focusing on product management and generative AI. By joining this course, you will get access to over 3 hours of HD video content. Step by step tutorials and activities, highlighting real world practical applications of Claude generative AI tools, PDF summaries for reviewing the key insights from the course and much, much more. We will kick off the course by learning what Claude is capable of how to communicate with Claude and structure your requests and how to make Claude work best for you. We'll go through a series of practical scenarios such as brainstorming ideas for your next project, drafting, marketing emails, summarizing long form content, creating mermaid diagrams for visual process descriptions, and even creating simple games without technical knowledge, all with Claude as your assistant we also learn some advanced techniques for solving complex problems with cloud, how to spot incorrect responses from AI, and most importantly, how to prevent them. And here is the exciting part. You don't need to be a tech wizard or AI expert to use Cloud AI. It's intuitive interface makes it simple for anyone to get started and see results in minutes without prior experience in AI, generative AI or programming. So let's begin the course by covering what Claude is, its main capabilities, and how people are achieving more with Claude Alca in the next medium. 2. Claude’s overview and main use cases: Hello everybody. Ever wished for a colleague who's available 247 never gets started and actually enjoys explaining things well met Claude today, we are exploring what Claude is, its main capabilities, and how people are putting this tireless assistant to work. Let's dive in. So what exactly is Claude AI? Claude is a family of large language models, LLMs for short, developed by a company called tropic. The name Claude honors Claude Shannon, a pioneering scientist whose work was fundamental to the development of artificial intelligence and information theory. What does a language model mean in simple terms? A language model is a type of artificial intelligence that's trained to understand and generate human language. Think of it as a highly sophisticated computer program that can process and respond to text in ways that feel natural and helpful. Unlike traditional software that follows rigid rules, Claude can understand context, engage in natural conversations, and adapt to your needs. What makes Claude special is its ability to communicate in a way that feels natural and genuinely helpful. By the way, if you would like to explore more about what EI is and generative EI in particular, don't forget to check the deep dive section in the course where we talk about these topics. Now let's look at Claude's main capabilities. First, Claude excels at understanding and working with language. It can help you write and edit text from emails to articles to creative stories. It's like having a writing partner who offers suggestions, refines your ideas, and catches potential improvements. Second, Claude is remarkable at analysis and problem solving. It can break down complex problems into manageable steps, analyze data and explain difficult concepts in simple terms. Whether you are studying mathematics, trying to understand scientific concepts or analyzing business data, Claude can guide you through the process Third, Claude is highly capable of assisting with programming and coding tasks. It can write code in various programming languages, help debug existing code, explain programming concepts, and even assist in building entire applications, whether you are a beginner, learning to code or an experienced developer tackling complex programming challenges, Claude can provide valuable support. Fourth, Claude has impressive research and learning capabilities. It can help you gather and synthesize information, explain complex topics, and answer questions across a wide range of subjects. Think of it as having access to a knowledgeable tutor, who can explain things in ways that match your learning style. However, it is important to note that as of the time this video was recorded, Claude does not have access to the Internet, and the cutoff date depends on which model from the Claude family of AAM's you choose to work with. So note that Cloud may not be fully up to date on events, developments or information after that date, and you may need to rely on other resources or language models for your research. Okay, and finally, let's talk about how people use clot in real life. On Tropic, the company behind Cloud has developed a system called Cleo, a privacy reserving analysis tool that helps understand how people use Cloude while keeping their conversations private and secure. It is similar to how Google Trans shows what people are searching for without revealing individual searches based on Cleo's analysis of 1 million real conversations, web and mobile app development is the most popular use, accounting for 10.4% of all conversations. Content creation and communication follow at 9.2% academic research and writing represent 7.2% of usage, while education and career development come close behind at 7.1%. Advanced AI machine learning applications account for 6% of all conversations. The data also shows significant use for business strategy and operations, language translation, DevOps and cloud infrastructure, digital marketing and CO and data analysis and visualization. Looking at these numbers, one thing becomes clear. Claude isn't just a language model. It is becoming one of the world's most resourceful colleagues and it never needs a coffee break. Let's recap the key points we've covered here. 3. Claude family overview. Creating an account and first interaction with Claude.: On everyone. In this lecture, let's learn what models are available within the Cloud family of large language models and how to get access to them. The clot family launched with three models, each named after different types of poems, Haiku, Sant, and Opus, just like poems, each has its own characteristics and best uses. Let's break them down one by one. First, we have clade Haikum. This is like the sprinter of the family, fast and efficient. It's perfect for quick everyday tasks like drafting emails, answering questions or helping with simple analysis. If you need rapid responses and don't require deep complex thinking, haiku is your go to choice. It's also the most cost effective option, making it ideal for businesses that need to process many requests quickly. Next up is clouded Sanet which sits in the middle of the family. Think of Sant as the rounder, it strikes a nice balance between speed and capability. Lastly, there is Cloud Opus, which is like having a seasoned expert on your team, while it might be a bit slower and more expensive than its siblings. It's the one you want for tasks that require deep analysis, sophisticated problem solving, or creative writing projects. However, the clot family has evolved dramatically since its initial release. Here is where we stand today. The current flagship model is Cloud 3.5 SNET released in October 2024. This newest model matches or exceeds the original opus capabilities. It shows significant improvements in code generation, 50% success rate on real world, programming tasks, reasoning capabilities, natural conversation, multi model understanding, and computer use abilities, which can do things like navigate websites and apps click buttons, type, read screenshots, and even follow multi step instructions to get tricky tasks done. What makes this so different from how we've worked with AI before is that we used to need special tools or connections, like figuring out a way for the AI to talk to a website to make things happen. Now Clote can just browse a website or use an app like you would no extra setup needed. This puts AI on a whole new level, letting it handle everything from boring data entry to complex tasks and making it feel like a real assistant rather than just an advisor. And Claude isn't the only one stepping into this space. Other companies like Open AI have released similar tools such as operator. However, both Clouds and OpenAI's computer use features are still in the early stages and not widely available yet. If you are excited about this technology, keep an eye out for updates as it develops further. Let's continue with an overview of Cloud's family of large language models. Clot Haiku has a new addition to the family as well, clot 3.5 Haiku, and it also achieves strong performance among models in its class. It demonstrates improvements over its predecessor and in many cases, performs comparably to the original clot 3.5 Sonnett and clothe three opus models. As for the knowledge cutoff dates, the knowledge cut off for the upgraded Cloud 3.5 SNET is April 2024, the same as that of the original Cloud 3.5 SNET model. The knowledge cut off for Cloud 3.5 Haiku is July 2024. Now with this knowledge in mind, let's jump straight to the subscription options. Currently, Cloud comes with the following four plans. You can start completely free of charge, see what Cloud is capable of, and then upgrade to one of the premium plans. The free plan comes with access to one of the latest models only usually cloud 3.5 sunt. However, during peak capacity, it might switch to Haikum. This model access is subject to usage limits that ensure fair usage across all users. Usage limits refer to the maximum number of messages you can send, which depends on many factors, including message length, the length of files you attach to the conversation, the length of the current conversation, and clots current capacity. The definition of usage limits may be a bit confusing at this stage, since we are just getting started with Cloude. However, you will understand the idea behind usage limits as soon as you start working with Cloude. We'll also discuss how to optimize usage limits and get the most out of your subscription plan. For now, remember that usage limits vary from one subscription plan to another with the free service offering the minimum usage and the pro plan providing at least five X the usage, compare it to the free service. Let's create an account with clot to sign up. Open clot at clod.ai. Next, provide your email. Click on continue with email. Clote says that you have to check your email for the verification code. This is a bit confusing since, in fact, instead of verification code, Clote will send you a link. You should click to continue your registration process. Let me copy the link. Come back to my browser I use for this demo and copy the link here. Next, you provide your phone number. Your date of birth to register an account with Cloude, you have to agree to anthropic consumer terms and usage policy. Next, let's hit on Sand verification code. Now the code will be sent as an SMS to the mobile number that you indicated on the previous screen. Let's hit verify and create an account. I already used my mobile phone to register my main account, so I cannot proceed further with the same number. However, in your case, you will be successfully registered with Claude and next, you will see the Claude main screen. The interface is intentionally minimal and clean designed to feel like a natural conversation. Think of it as opening a fresh notebook, simple, uncluttered, and ready for your input. On the left side of your screen, you will see your conversation history. Each pass chat is neatly organized here. Similar to how email threads are arranged in your inbox. You can easily click on any previous conversation to revisit it. The main chat area takes up most of your screen. This is where the magic happens where you and Claude interact. You will see a simple message box where you can type your questions or requests. And while we are here, let's explore several tools available at the bottom of the chat section. First, you can choose model you'd like to use for these conversations. If you are registered for the free version of the account, you will have access to clothe 3.5 Sant only. However, if you are registered for a paid subscription, can choose from one of the three models prior to starting your conversation with Claude. Next, you can choose a response style that Claude should adopt when answering your questions or completing a task. We will learn how to work with these styles in the subsequent sections of the course. Finally, if you'd like to start a new chat, go to the left hand side menu and click on Start New hat and from here, type in your message. Okay. And that's it for this video, we are almost set to start communicating with Claude, but this lecture would not be complete without a brief summary of what we've just covered. An 4. Getting started with Claude: Section Intro: Welcome to the new section on Cloud AI. In this part of the course, we are diving into the exciting world of prompting the art of effectively communicating with Cloud to get the best results. We will start by breaking down the basics. What is a prompt? What is a prompting? And how does Prompt engineering fit into all these? Plus, we'll touch on concepts like chat prompton and enterprise Prompton you will know to tailor your approach for different contexts. Next, I will share a simple prompting formula to help you talk to Claude in a way that's clear and effective. From there, we will explore iterative prompting, building on Claude responses to refine and get even better results. You will also learn how to make Claude work best for you by setting up profile preferences and adjusting conversation styles to match your needs. Be practice makes perfect, I've included a few follow along exercises where you will brainstorm with Claude, fine tune prompts, and even create a marketing campaign together. We'll also cover essential skills like sharing content with Cloude using examples to guide it and formatting output to meet your goals. And by the end of this section, you will feel confident and ready to use Cloud AI for work or personal tasks. So let's get started. 5. What is prompt, prompting, prompt engineering, chat, and enterprise prompting.: One. Think of the last time you asked someone a question. The way you phrased that question likely influenced the answer you received. That's exactly what we are seeing today in the world of AI. We will start by breaking down three key terms that are essential to communicating with AI systems. What exactly is a prompt? What do we mean by prompting? How does prompt engineering tied together? We will also explore the distinction between chat and enterprise prompting. Let's get started. Prompt is the actual text or instruction you write to cloude or any AI model. It's like a message or queriu. The specific input you provide. Think of it as what? The actual content of your request, prompting is the act of writing these prompts. It's the general activity of interacting with and giving instructions to AI models. This is how the process of communicating with the model. Prompt engineering is a more specialized and systematic approach to creating and refining prompts. It may involve understanding how the model works, testing and iterating on prompts, considering edge cases and more. Think of it like cooking. A prompt is like a single recipe. Prompting is like cooking in general. And prompt engineering is like being a professional chef who systematically develops and tests recipes while considering ingredients, equipment, user preferences, and so on. Now, there are two main types of prompton you need to be aware of Enterprise promptin and chat prompton. Enterprise prompton refers to designing prompts for business applications where the prompts will be used repeatedly at scale. These fronts are engineered to handle diverse user inputs, maintain consistency and operate reliably within specific business constraints and requirements. They typically power customer facing applications or internal business tools. For example, a company may have a customer service assistant chat board designed to provide immediate 247 support for customers on the company's website and app the assistant may handle common technical issues, basic product inquiries, and routine tasks like processing returns or refunds, enterprise prompton will be used to customize how the assistant must reply to a customer depending on their request. Such prompts will be used many times, 1 million, 10 million, or even hundreds of millions of times, they need to be very reliable and consistent account for real user behavior, typos, unclear requests, and so on. And handle a wide range of user inputs and etch cases. Chat prompting, on the other hand, refers to direct conversational interactions between humans and EI models in chat interfaces for immediate specific tasks. This type of prompting is typically more flexible and informal, allowing for real time interaction and clarification through dialogue. It doesn't need to handle as many edge cases as enterprise prompton and can be defined through conversation. For example, using clod.ai to help write an email or analyze a document would be chat Prompton. Chat prompton is fundamentally different from enterprise prompting. And in this course, we are going to cover chat prompting using cloud.ai. Why do we talk so much about this distinction between enterprise and chat prompting? Well, because as we just discussed, the way we structure and refine prompts will be different depending on whether we plan to use the prompt for enterprise or chat settings. If you research additional materials on prompting and prompt engineering, including those from anthropic, you might come across quite a lot of resources covering how to structure enterprise prompts. However, if you plan to use Cloud mainly through the chat interface, this information is not something you can benefit from. So keep this distinction in mind and don't spend time diving deep into enterprise prompting if your primary use case is chat based interactions. All right. And now that we are on the same page with the terminology, let's dive straight into the nitty gritty of chat Promptin. I'll see you in the next lecture. 6. How to talk to Claude: Prompting formula: Everyone welcome to our first lecture on chat prompting. Here, you will learn how to approach creating and refining prompts that can be used in the clade AI chat interface, as well as with other similar models. Let's get started. When chatting with the friend, you don't use rigid templates or formal structures. You have a natural flowing conversation. The same principle applies to chat prompting with AI models. However, there are times when a bit of structure can help us get better results and make one prompt more effective than another. So let's cover the key ingredients of an effective prompt. The central part of every prompt is the core intent or task. This can take the form of instructions such as write a five paragraph email to introduce a new productivity app to small business owners, focusing on its time saving features. Think of instructions as the task. You want the model to perform. Another form the intent can take is a question such as, what steps should I follow to create a compelling incident profile? How do I structure a business plan for a startup idea? When writing a task, your goal is to be clear and specific about what you'd like to achieve. Writing something like, help me with the presentation won't be enough to get a high quality document. You can confidently present to your boss, colleagues or investors. As rule thumb, remember that anyone without specific knowledge of your subject should be able to understand your prompt and execute on it. If they would be confused about how to follow your instructions, Claude will be confused as well. Don't assume Claude has any contextual information about your task such as how the results will be used, who the intendant audience is, what successful task completion looks like, or a list of points you won't cover it. You need to provide these context or task details yourself. For example, if you want to create a presentation, include information about the number of slides, the purpose of the presentation, the key topics to be covered. Here is an example of a well crafted prompt. Create a seven slide presentation on the topic of personal branding. Include what it is, white matters, key components, and steps to develop your brand. Another example, explain how to write a compelling email in five easiest steps. The instructions should cover crafting and engaging subject line, structuring the email clearly and using a professional tone. Make the process simple enough for anyone to follow, even without prior experience in formal writing. You can provide context, not just for the task itself, but also for the tone you would like Cloud to use. For instance, use a conversational tone that balances professionalism with accessibility. You can also specify rules or constraints Cloud should follow. For example, in the email writing guide prompt above, you might add. Here are some important rules for writing the explanation. Keep each step explanation 2-4 sentences. Provide at least one do and one don't example for each step. Incorporate formatting tips like spacing paragraphs, bullet points. Avoid technical jargon or complex business terminology. Another way to enhance your prompt is to assign Cloud a specific role when performing a task. This is also known as role prompting. Role playing helps AI models adopt the nuances of specific perspectives, improving the relevance and quality of their responses. For example, act as a seasoned executive assistant with over 15 years of experience managing high level business correspondence or pretend to be a professional writer, turned email writing consultant. I also came across a clever recommendation to introduce a role as being the world's leading expert in whatever I'm about to ask you about. While this can improve performance, I've personally found that specifying a well defined role tends to get better results. I encourage you to test this prompt yourself and share your results in the Q&A section for this video. You can take role prompting a step further by providing audience context in addition to the role. For instance, as a senior executive assistant with 15 plus years of experience managing high level business correspondence, create a guide for software engineers and other technical people looking to improve their business andmil communication skills. Notice how Claude changes the examples for dos and don'ts to make them relatable for technical professionals. It's pretty amazing. If you are feeling overwhelmed by the idea of crafting such detailed prompts, don't worry. The beauty of working in a chat interface is that you don't need to design a perfectly thought out prompt to begin the conversation. You start with a broad question or task and refine it through dialogue with the AI model. This iterative approach allows you to clarify your needs and improve the response you receive over time. We will talk more about the iterative prompting in our next video, and for now, let's sum up what we talked about in this lecture. The central part of every prompt is the core intent or task, which can be expressed as an instruction or a question. Providing context, tone, and rules ensures that prompts are clear and specific, making it easier for AI to generate accurate responses. Role prompting involves assigning the AI a specific role to adapt, improving the quality and relevance of its outputs. Including audience context helps tailor the AI's responses to the need of a specific group or demographic, chat interface allow for iterative prompting, helping users to refine tasks and responses through ongoing dialogue. That's it for this video, ACA in the next one. 7. Building on Claude’s responses: Iterative prompting: Everyone. Welcome back. If after watching the previous lecture, you feel like creating a good prompt is an arduous task and that you need to turn into a prompt engineer to succeed in this job. Here is a secret the experts use. Think of prompting as a conversation or a multi step process, not a one time question. Just like you might clarify directions in a new city with a local, you can refine your prompts based on close responses. Let's walk through a real world example of iterative prompting to see how it works. Let's say we would like Claude to help us create a business proposal for a mobile dog grooming service. Step one, the initial prompt may be quite broad like create an outline for a business proposal for a mobile dog grooming service. In the second step, we narrow down or refine our initial request by asking cute something like take the outline you created and expand the market analysis section, focus on demographic data and competition in urban areas. A On the third step, we ask for specific details. For instance, now develop the financial projections section, include startup costs, monthly operating expenses, and revenue forecasts for the first year. We can repeat step two and three several times depending on how satisfied we are with the responses. Please note that just like a skilled project manager builds on previous discussions and decisions, Cloud maintains context throughout your conversation, allowing you to reference and expand on earlier points in your interaction. This is a technique called memory referencing. You might ask something like, remember the marketing strategy we discussed earlier. Let's build on that, but focus on suburban families in areas with limited grooming options. Of course, if you feel that your conversation is not going in the right direction, you always have the option to start over and reframe the very first question. The final step of the iterative process usually involves asking Claude to polish the response, review the entire proposal and enhance the executive summary to highlight our unique value proposition and market opportunity. Alternatively, you can ask Claude to provide feedback on the entire piece of content. In this case, the business proposal, focusing on how it can be further improved. Then you can include those changes in the final version of the document. A this step by step approach allows you to review and refine the output at each stage. Make adjustments based on intermediate results, maintain control over the final product and build complexity gradually. Think of it like sculpting. You start with the basic shape and gradually refine the details until you achieve exactly what you want, and that's it for this video. Let's sum up the key points that we've just covered. 8. Making Claude work best for you: Profile preferences and conversation styles: Hello, and welcome to the new video. We are going to explore how to make clothe truly yours by using two powerful customization features, profile preferences and conversation styles. Whether you are a student, professional or casual user, understanding these features will help you get the most out of your interactions. Let's start with profile preferences. When you set up your Cloud account, you can customize various settings that affect how Cloud interacts with you. Think of it like setting up your smartphone the settings you choose will create your ideal working environment. To open profile settings, click on your user name at the bottom of the left hand bar menu and from here, choose settings. Here, you can first set your name, specify how clothes should address you and indicate the field you're working in. Next, you will see a section for setting up personal preferences. There are several things you can configure here. First, is contextual preferences. Information about your background and needs, including your role, your area of expertise, common tasks, you work on approaches or methodologies, you like to use your audience or who you work with, your goals for using Cloude. Second, is behavioral preferences. So how you want Claude to respond. This includes such things as communication style and tone you prefer output, format preferences, level of detail needed, language preferences, and how you want information presented. For instance, this is the description I have in my profile preferences. I first said that I'm a product builder and online instructor. I said that I frequently create educational content and lecture scripts on software, product management, AI, and generative AI. Next, I specified my target audience, saying that typically it includes business people, product managers, and non technical stakeholders. And I next specify how I use clothe AI, namely for research, brainstorming, and writes and lecture scripts for my courses. And next, you'll see a list of my behavioral preferences, which is quite extensive here I specified such things as I prefer breaking down complex technical concepts into simple understandable language. I prefer using real world examples and case studies to explain abstract concepts. I also prefer focusing on practical applications rather than theoretical details and so on. You're welcome to go through this list and take some preferences that you feel are applicable to you as well. Now let's talk about styles. Style selection is available at the bottom of the chat field. Styles are like different personalities. You can switch between depending on your current task. For example, if you are working on an academic paper, you might select a more formal style. If you are brainstorming creative ideas, you could switch to a more casual style. The beauty of styles is that you can change them anytime during your conversation with Claude. Let's see how it works. Practically speaking, given that the majority of the work you will do with clothe will belong to a certain domain like educational content creation in my case, you will use the same style most of the time. I found myself using the normal style more often than others. And it is selected by default when opening a new chat. And if I need to tweak the style a bit for a specific chat, I would rather include specific instructions in the prompt than modify the style itself. But this is my work routine, which may not necessarily work for you. So definitely check out the styles to find those suitable for your needs. Perhaps you have already noticed that you can also customize your own style, a feature that helps reproduce your unique writing voice and style. Since we are just beginning our experiments with Claude, I would not recommend customizing your own style just yet. Instead, focus on experimenting with models and default styles and notice which ones work or don't work for your projects. Once you are familiar with Claude standard style settings, you're welcome to join me at the lecture dedicated to creating custom styles, which will come in the subsequent sections of the course. And that's it about profile preferences and conversation styles. I'll meet you in the next video. 9. Follow-along: Brainstorming with Claude: Okay Let's begin our experiment by using a very short prompt, like, give me some ideas for a side hustle project. I'll use the newest clothe 3.5 sonnet for this demo. We see that even with this short prompt, I've been able to get some initial ideas that are relevant to my professional domain. This is because I filled in my profile preferences with information about my background and what I do on a daily basis. Some of the ideas are really great. These are projects I would seriously consider if I decided to run a side hustle for real. But let's revisit our prompt and see what we can do to improve it. I'm definitely missing some context for the task I want Claude to do. I would add more details about the types of projects I'd like to work on. First, I would specify that I want them to focus on my core expertise, product management. This is to ensure Claude does not include projects from unrelated domains. I would also mention that I have limited time to dedicate to this project since it is just a side hassle. Lastly, I would specify that I want the project to be profitable and I would include my target earnings. Why I mention all this? Because these details are relevant to the project ideas I'm researching. And I believe giving Claude this context, we get better results. Lastly, let's also add details by highlighting a list of topics. I want to be covered in the response. For starting a new sentence on a fresh line, press Shift and return if you are on a Mac and shift and enter if you are on Windows. These details provide specific parameters for the brainstorming session, including the number of ideas, their implementation steps, and possible monetization strategies. This results in a more structured and useful output. Let's submit this prompt and see if we get better results. Actually, let me open a new chat to ensure that results from the previous conversation do not interfere with the new prompt. I'll copy this text and paste it to a new chat. And let's hit Enter. Here are the results, pretty good. They are definitely more detailed and well thought out than those from the first iteration. And in case if you are not satisfied with the list of ideas, you can ask Claude to propose ten additional ideas. I've noticed that sometimes when you are brainstorming and not getting creative interesting ideas, it can help to ask Claude for new variations. Not just once, but three, four or even five times. Occasionally, you will receive brilliant suggestions through these iterations that you wouldn't have gotten otherwise. Let's try to do this. Okay, great. We've got 30 different ideas we can choose from. But before we proceed, let's also include a role for clade to play at the beginning of our prompt. I'll copy my original prompt, open a new chat. And I would add this role at the very beginning of the prompt. You are an expert in brainstorming techniques with over 15 years of experience. I haven't changed any other details of our previous prompt, so let's hit Enter and see the results. Okay. Great. I see several ideas that I really like and I can take them as a side hassle like this one, product management productivity tools. But before we go ahead researching more on these ideas, let's do one more experiment and substitute this role description with another one. I'll copy the prompt. Open a new chat. Let's remove the asterisk. And instead of this role description, I'll include another one. You are the world's leading expert in whatever I'm about to ask you about. Yes, it's a funny role description, but nevertheless, let's test if it can get us better results. Great suggestions as well. But frankly, I don't see any significant changes if we compare these results with our previous iteration. So you can experiment with including this role description for your proms and see if it can make a difference for your use case. 10. Follow-along: Ask Claude to improve your prompt!: One. Welcome back. Before you start practicing brainstorming with Cloude, let me show you a quick technique you can use to enhance your prompt, especially at the beginning of your experiments with Clote When you're just learning prompting techniques, you can ask Claude to help you improve your prompt. To do this, open a new chat. Type in your request followed by the prompt description. I include the prompt text in quotes to indicate where the prompt begins and ends. Let's press Enter and see what clades response is. We've received quite a detailed enhanced version of the prompt. Of course, you don't have to include all the details from the original suggestion. For instance, some parts might not be relevant to the project ideas, I'm brainstorming. Use this prompt as a general guideline for what to include, but be sure to adjust it for your specific use case. Let's make modifications to the prompt that Claude suggested to us. The easiest way to make changes in this prompt is to first copy the entire text from the conversation by pressing copy. Then you can open a new document. For me, it is a Google Drive document. I copy paste the text here. You see that we have the prompt plus some additional text with close information on the changes and the improvements that he made to the prompt. So I'll delete all the parts that do not belong to the prompt. Okay, and now we can make all the changes that we'd like. So I just made one small change by adding risks to the opportunity description, and I think I'm fine with all other details. So let me copy this text and paste it to a new chat with Claude. All look great, and let's hit Enter. Okay, pretty nice job. I found that this new information can definitely be helpful when developing these ideas further. However, I cannot find the information about project challenges and risks, even though I have requested this information from Claude. So let me ask Claude to provide this information for each of the three ideas. Yes. Great insights so far. And I found that this new information about project challenges and risks, something I hadn't thought of before, is very helpful for assessing the viability of site hustle projects. And what is interesting here is that apart from providing direct response to my question, Claude also gives us suggestions about common risks across all opportunities. Of course, I can continue talking to Claude and ask any additional questions regarding the three opportunities that we just saw, or perhaps I can ask Claude to give me other ideas I can consider. And that's it for this quick tutorial. I hope you like this technique of asking Claude to improve your prompt and you will start using it for your projects and Ilsa in the next video. 11. How to share content with Claude: One. Welcome back. In the previous lectures on prompt Engineering with Claude, we talked a lot about how to frame your instructions and what information to include. However, apart from the instructions themselves, oftentimes you may also need to submit certain documents that need reviewing and analysis. Let's see how it works. You can submit the information from the documents you want clot to act on directly in the prompt field, or you can attach the entire document to your chat. The first option works well when you need to work on a specific textual fragment of your document. For instance, if I want clod suggestions on a particular part of my resume and not the entire document, I would opt for submitting this fragment directly to the chat like this. However, oftentimes you need Cloude to work with the entire text document, or you might have a PDF file or Excel spreadsheet. You need help analyzing. For these cases, you can upload a document into your chat. Clote can work with many different types of files, including PDFs, word documents, Excel spreadsheets, CSV files, and plain text files. Uploading a file is straightforward. You can choose from three different options. You can upload a file from your local drive, or you can take a screenshot, or you can upload a file from your Google Drive. And of course, you can just dragon drop document to the chat section. Once the document is uploaded, you will see the file appear in your chat. Now here is something important to remember. Claude can see the entire content of your file just like you can. This means you don't need to copy and paste the content. Claude already has access to it. However, you do need to tell Claude what you wanted to do with the file. For example, if you've uploaded a spreadsheet, you might say something like, can you analyze the sales data in this file and tell me the top performing months, or if you've uploaded a research paper, you don't want to read yourself, but want an executive summary, you could ask something like Could you summarize the main points from pages 3 to five of this PDF? Notice how specific I was in my request. I didn't say something generic like, what do you think about this file? The more specific you are, the better Claude can help you. All right. Let's talk about working with multiple files. Yes, you can upload several files in one chat. Claude can compare documents, cross reference information, or work with related files together. For example, you might upload two versions of the same document and ask Claude to identify the differences between them. Oftentimes, you need to tell Claude which file you're referring to. Think of it like having several documents on your desk. You need to point to the specific one you want to discuss. The simplest way to reference a file is to use its exact file name. For example, if you have uploaded two CSV files, you could say something like this. Please compare the first quarter sales in sales 2023 CSE with those in sales 2024 CSV. You see that clade tells me that the file size I'm using here is over the limit. So to continue the conversation, I have to revisit the documents and see what I can do to reduce their size. When working with three or more files, you might want to number your requests. Let's say that we need a comprehensive software development life cycle analysis across our project documentation. The goal is to track software requirements from initial specification through implementation to testing, identifying any gaps, discrepancies, or quality issues in our process. This analysis will help ensure our software meets all specified requirements and quality standards before deployment. We can attach the following three files into the chat. And ask Claude to analyze them in this order. First, read the requirements from specifications dot dog. Then check if these requirements are met in implementation dot PDF, and finally, list any discrepancies in comparison with testing results CSV. And by the way, if you are going to reference the same files multiple times in your conversation, you can establish short nicknames at the beginning. Just say something like, I would refer to quarter forecast 2021, CSV as the forecast file and actual 2024 as the actual file throughout our conversation. Lastly, please remember that while Claude can read your files, it cannot modify them directly. Instead, it will provide you with suggestions, analysis, or new content that you can use to update your files yourself. That's it for the lecture. Let's briefly sum up what we've learned here. Claude accepts common file formats including PDFs, Word documents, CSVs, text files, and others. Files are easily uploaded using the upload button in your chat interface. You need to give Claude clear instructions about what you wanted to do with the files. Being specific with your requests leads to better results. You can upload multiple files and ask Claude to work with them together. While Claude can read and analyze files, it cannot modify them directly. That's it for this lecture and Ilsa in the next one. 12. Using examples when prompting: One and welcome back to the new lecture where we continue talking about how to communicate with Claude and what to include in your prompt description. So far, we've covered several components that can be included in a prompt a task or what you'd like to achieve, followed by specific details or context and rules necessary to perform the task or answer a question. Next is role context, a specific role that Claude will be playing when performing a task. Optionally, you can also introduce the intended audience for your task. Lastly, we mentioned that you can share additional content with Claude by attaching documents to your conversation or by including the text as input data directly in the chat and regarding the ordering of components in your prompt, the ordering matters for some elements, but not for others. For instance, it is recommended to include roll context earlier in the prompt, while input data might not be necessary depending on the task and its ordering is also flexible. But in general, if you stick to the ordering shown in the course presentation slides, it will be a great start to an effective prompt. Okay, let's introduce another prompt element. Examples also known as shots act as demonstrations that guide the generative AI model on what kind of output you are looking for, including the answer format and what you want to avoid. Perhaps you've heard of terms like one shot or a few shot prompting. These refer to using one or several examples in your prompt description. For chat prompting, examples typically demonstrate tone, like formal versus informal, serious versus schedule, empathetic versus matter of fact and style such as sentence length, format patterns, bullet points, versus paragraphs, technical details level, basic or advanced terminology, and so on. Let's go over some concrete examples. So in the scenarios you just saw, we used examples to demonstrate both style and tone for the desired response from Claude. Remember our previous lecture example of an email writing guide. We ask the AI model to use conversational language that balances professionalism with accessibility. It turns out you can achieve similar results by using different prompting techniques. If it is easier to provide an example of the output you are looking for rather than giving a detailed description. By all means, do so. Apart from one or few shot prompting, there is another technique using interactive examples. Interactive examples differ from regular examples in that they can create a dynamic back and forth learning experience where each example builds upon previous understanding or feedback while regular examples are static demonstrations. Interactive examples involve active participation and iteration. Here is how interactive examples work. You provide an initial version example. The EI gives specific feedback and suggestions. You create an improved version based on that feedback. I The AI analyzes the improvements and suggests further refinements. You iterate again if needed. The key is that each iteration builds on the feedback from the previous version, creating a collaborative improvement process. Okay, great. And that's it for this video. Let's quickly cover what we've just learned here. 13. Output formatting: One. We are almost done covering the key ingredients of a good prompt. There is yet another component you may find worth including in your prompt, information on what format you want clots response to take. Let's talk about this now. Remember that in our first lecture on prompting, we said it's important to include information regarding the basic outline or list of points, you won't cover it as context for your task to clot. It turns out you can also specify your formatting preferences for close response, which can help organize information more effectively. This information may not be necessary depending on the task, but if you include it, adding it toward the end of the prompt is better than at the beginning. Let's go through some examples of formatting you can request. You can ask for specific formatting styles. For example, if you need a business report, you might say, please format this as a professional report with headers, subheaders and short clear paragraphs. Clot will structure the information accordingly, making it ready for professional use. When working with data or analysis, you can request tables or specific layouts. Instead of a wall of text, you might say, present the comparison of these three products in a clear table format with features in the left column. This makes complex information easier to understand and use. You can request specific markdown formatting. Claude can use bold text, italics, headers, and bullet points as needed. Just ask for key points in bold or important terms in italics, and Claude will do its job. You can organize your tips using bullet points for clarity. Min tip, supporting detail and another detail. Lastly, remember that you can always ask Claude to reformat its response if the first version isn't quite what you needed. It's perfectly fine to say, could you reorganize this information as a number at least? Or please break these into shorter paragraphs for better readability. That's it for this brief lecture, let's recap the key points we've just covered. 14. Follow-along: Creating marketing campaign emails (part 1): One, welcome to the new follow on lecture. Here we are going to explore a use case, I believe is one of Clothes strongest. We will create marketing materials, specifically marketing email sequence, which will be used to spread the word about a new product among prospective customers and invite them to try its trial version. Was the first scenario I used Cloud four and I was impressed by the quality of the results. I decided that I definitely want to have Cloud AI among my generative AI tools. Here is some information. I prepared for the tutorial. The company name is narrative systems. It is a forward thinking AI software startup headquartered in Austin, Texas. The company specializes in developing enterprise grade generative AI solutions that help businesses automate and enhance their creative processes. The product that narrative systems is about to launch is called slide symphony. It is an innovative presentation generation platform that transforms text documents and verbal descriptions into polished professional presentations. The system understands context, hierarchy, and narrative flow automatically creating and visually engaging slides with appropriate layouts, graphics, and data visualizations. In addition, here is the current version of the email pitch created by one of the company's software engineers. The purpose of the email is to share information about the product and its value and to invite prospective users to join the trial version. Usually such emails are prepared by the marketing team, but narrative system is still a small startup with just a handful of team members working on launching the first product. As a result, there is no dedicated marketing professional on their team. Unfortunately, after sending this email to its list of contacts, the company did not receive enough attention. Just a few people clicked on the trial offer and only one actually signed up for it. Let's see if Claude can help us improve the situation. So I just opened a new chat and here is my first request to Claude. I said that I want to get feedback on the email pitch for my new product. I explained the problem that I currently have with that email and I ask Claude to provide his opinion on possible issues with my current email. Next, I said that I will submit the text of the email and I expect Claude to provide feedback. Notice that here I'm using a technique often called task framing or two step prompting. The first message that you see here on the screen sets the stage for what I'd like to do explains the context and requirements of the task and confirms Cloud understanding of what is required. I'm going to submit the text of the email page in my next message. This approach typically leads to more thorough and target feedback, compare it to providing everything at once. Let's hit Enter and see what Claude replies. Great Claude acknowledges our request and is looking forward to work with my email. Let's copy this text and paste it into the chat. I don't need to provide any details here, as I already stated my request in the first message, I'll hit Enter one more time. See that Claude performs an analysis of my current email saying that it has several issues, including the fact that the subject line is too generic, that it now has a lot of technical details, which is, in fact, true, since as you remember, this email was created by software engineer and Claude also says that the email is to feature focused, completely agree with this and it is missing an emotional appeal. Here is the suggested structure of a new email. Et's reply to Claude. Here I'm using interactive confirmation by saying to Claude that the revised email looks awesome. Doing so helps Claude calibrate its responses. For example, if you say, yes, that's right, but I am especially interested in X, Claude can adjust its focus accordingly. Alternatively, if you say something like the revised email doesn't look quite right, I'm actually looking for a technical accuracy review more than marketing effectiveness. This would completely reshape Claude's approach to reviewing your email. After saying that I like the first email that Claude provided, I also asked Claude to create the extended marketing package, which will include several options for an email sequence. I specifically asked for three emails in each sequence, and I said that I will use those emails to maybe test the sequences to see what works best. All right, Claude replied that it created three distinct email sequences, each with different focus. Sequence A focuses on describing problem and solution. For sequence B, we have a focus on feature and benefits for the sequence C, we have a time saving focus. In these variations of sequences that take different psychological approaches can provide fresh insights for those not familiar with various marketing strategies that can be used to evoke a certain feelings in people and nudge them to try out your product. I would definitely consider the sequences for the marketing campaign we are working on now. Let's also ask Claude to create a second variation for all the emails. I really like that clot at some images, icons and symbols. I think this is a really great addition to the text. All right. The new versions of emails look great. To decide how to proceed next, we need to review them carefully, probably together with other team members. Let me share my experience which might be useful for some of you. When I received similar emails from Claude for the product I was about to launch, I realized that the text in the emails was still quite shallow and didn't explain the problem the product solves or its value proposition clearly enough to attract the right users. It wasn't because Claude didn't do the right job. Close email suggestions were, in fact, great. The issue we faced back then was that we hadn't formulated the problem, the solution, and how our product could address that problem before starting to prepare the marketing emails. It was a missing part. The insights we needed to generate first before attempting to create a great marketing email sequence. Here is what we did. We got together as a team and formulated all the key messages regarding the problem, the solution, and our products value proposition. Then we took the emails Cloude created, similar to the one that you currently see on the right hand side of the screen. We modified those emails by adding that extra information from our brainstorming sessions, and then we gave the revised emails back to Cloude for feedback and further improvement. In the second part of this tutorial, I'll show you how to ask Claude for feedback and generate the remaining emails in the sequence. You'll also practice brainstorming a subject lines for the emails. Ilsa in the second part. 15. Follow-along: Creating marketing campaign emails (part 2): One and welcome to the second part of the tutorial, where we look at an example of using Clote for marketing, specifically for preparing marketing email sequences, for prospective clients, informing them about new product launch and inviting them to join the products free trial. Here is an example of a new email that I created based on the first variation provided by Claude. Here I expanded a bit on the problem that my clients might face, as well as solution that I suggest. Now let's come back to Cloude. I'm in the same conversation that we created in the first tutorial. Let's continue talking to Cloude here and now I'll ask. I rewrote some of my emails and now want your feedback on these new variations. Are you fine with this? I again use the task framing technique that we already covered in the first tutorial for this demo. Next, I'll copy the version to Emil beach and paste it into Clote chat. By the way, I often get questions about how important it is to write prompts with correct grammar. Afterall, Claude can understand messages with imperfect grammar. In case I need to get something done quickly, I can just rush to type my prompt without caring much about grammar. That said, clear and grammatically correct instructions do help ensure more accurate responses. However, it's not as critical as being clear about your intent. So describing clearly what you want Claude to do is far more important. Now let me press Enter and see what cloud feedback is going to be. As always, Claude gives us feedback and a new version of the email which we can review, modify if needed, and then ask for Claude feedback again. This is a great instance of using interactive examples when instead of asking for improvement ones, you go back and forth with Claude, refining your work more and more. And since we are creating an Emil sequence, let's ask Claude to create the second and third emails. For the second email, I would like to focus specifically on the quality of my product, and I give this information to Claude. I also submit several paragraphs of text describing the product's quality. Since our sequence consists of three emails, let's ask Claude to create the third email. Here is how my instructions look like. Again, I'm using a two step prompting technique, which I really like, as you may guess from this demo. Let's see if Claude acknowledges my request. And I'm submitting the clients testimonials. So my idea for the third email is that I want Claude to take the testimonials from some of my clients and create the third email in the sequence confirming that my past clients use the product and think it is of great quality and value. So I copy and pasted the testimonials into the chat and let me press Enter. Alright, cool. I think we've got some great ideas from Claude on how we can structure our email in the sequence. And now I suggest that we move on with this demo, and the last thing that we can do here to complete the marketing email preparation is to ask Claude to suggest the best subject lines for the emails. Let me ask the following. I think that I'll continue speaking with Claude in the same conversation in case it might need a context about the company and all other details that we discussed before. Next, let's submit an email text. Let's take the email from the third sequence. Let's say this one. And here is the clouds response. If we want to get more suggestions, we can ask. And if we don't like these alternatives either, we can ask something like yes. I found that if you're brainstorming ideas and don't get good options from the first iterations, it might be helpful to repeat the process a few times. Occasionally, you can get great suggestions. Let's do a few more iterations. So here I give the specific feedback on what part of my product offering, I'd like to focus on in the subject line. Alright, I think you got some great examples for how you can brainstorm a subject line for your email or other piece of content that you might want to create with Claude. And that's it for this demo, I hope you had a chance to notice the value Claude can bring as your AI marketing assistant whether you are coming up with fresh ideas, tweaking your copy or just need a helping hand, clothes got your back. Give it a shot and see how it can take some of the workload of your plate, and as always, Alca in the next video. 16. Understanding Claude's artifacts and projects: Section Intro: Welcome to the new section of the course where we will explore some of Claude's most powerful features, artifacts and projects. First, we'll explore artifacts, what they are and how Claude uses them. You will also learn how to maximize your clade usage limit with the help of artifacts. To put theory into practice, we will have a follow along exercise where you will create a process flow diagram step by step. This will give you a hands on experience in creating and modifying artifacts in the form of visual content. Next, we'll explore publishing and remixing artifacts and even remix a Tetris game. No coding involved. Finally, we'll cover projects, including how they keep content organized and how they complement artifacts. Are you ready to start? Let's jump in 17. Artifacts: Definition and how Claude uses them: Welcome everyone. In the first lecture of this section, we are going to talk about Claude artifacts, but don't worry. No ancient pyramids here, just powerful tools to help organize and structure your conversations with Claude. While you have already come across these artifacts in our previous section and maybe created some yourself, we are now going to formally introduce what they are and how they can transform your workflow let's start with what artifacts actually are. Imagine you are working with the designer. Instead of sketching directly in your notebook while you talk, they use a separate canvas to create their designs. This separate canvas allows them to focus on the artwork while maintaining a clear conversation with you. That's exactly what artifacts are for Claude. They are separate spaces where Claude can create and organize specific types of content while keeping your main conversation clear and focused. Why were artifacts created? Before artifacts, all of Clade outputs, whether it was code, stories or analysis, appeared directly in the chat. If someone was working on code for a website, remember, web development is the number one cloud use case. They had to copy paste the code into a separate file and then open this file in our web browser just to visualize the design clod created. This back and forth process obviously took a lot of time until one of Claude's team members came up with the idea of side by side interface where you can see the code or text on one part of the screen and the visualization on the other. That's how artifacts began. Apart from code, artifacts are handy for generating substantial piece of content like long stories or detailed analysis, mermaide diagrams, vector graphics, or even simple games. Artifacts appear next to your chat, letting you see, iterate, and build on your creations in real time whenever you need them. The great news is that artifacts are available on all clothed plans. To enable artifacts, navigate to your profile settings by clicking on your initials in the lower left corner and select settings. From the profile page, turn the enable artifacts to go on. Now let's go through an example of creating and modifying an artifact. I'm planning a relocation to Melbourne at the moment of recording this tutorial. So here is the real world example of my conversation with Claude, which involved creating artifacts. I first provided some context on the task I wanted Claude to perform. I need help researching secondary schools and ranting options in Melbourne, Australia. Then I stated the task. I needed the following information. For every school, provide the name of the district where it is located and the minimum and average rental prices for a house with three to four bedrooms and two bathrooms. I also submitted a list of schools as a file attachment. It turned out that there was an issue with this file, which Claude identified. So I resubmitted the list of schools in my second message. And here is the output in the form of an artifact. Claude organized the schools by geographic regions and provided rental estimates for each areas. As the next step, we might want to ask about additional details to include in the Melbourne Schools and housing analysis. I asked Claude for additional details such as typical commute times to the CBD, school zone boundaries, local amenities, and whether there is a park nearby. Notice that here, I'm using an interactive conformation technique we talked about earlier, acknowledging that I'm satisfied with the response Claude provided. Cloud can update an existing artifact in response to your message. The artifact window will update to show you the latest content. These edits, however, won't change Clote' memory of the original artifact content, and you can switch between versions using the version selector at the bottom left of the artifact. However, in my case, we see that a new artifact was created because I requested major changes affecting most of the content. By the way, you can open and view multiple artifacts in one conversation using the chat controls to access this, click on the slider icon in the upper right corner, select the artifact you'd like to reference and then continue where you last left off. Finally, you can make targeted updates when small changes to specific sections of an artifact are needed. In this scenario, Claude can update just the portion of the artifact while leaving the rest unchanged. To make targeted changes, select a sentence or a phrase where you want modifications to be made, and then click on Improve. Describe what you would like Claude to do, such as include five coffee shops in the list and click on Update. And here are the changes. Pretty nice job of Claude. Now, let's say we want details about several coffee shops from this list. So let's highlight one of them and click on Explain. Clot will provide a detailed breakdown of the information on the selected place, including its venue and space, information about menu, location benefits and so on. That was very quick and convenient. But please remember that Cloud is not connected to the Internet, at least at the moment of recording this tutorial. So if you request information subject to frequent changes, take time to double check it with another source. All right, we are all set with this tutorial, and I'll see you in the next lecture where we will talk about how to maximize your Cloud usage limits using artifacts. I'll see you there. 18. How to maximize your usage limits using artifacts: One and welcome back at the beginning of our course, we mentioned that we will discuss how to optimize usage limits and get the most out of your subscription plan. This is where using artifacts can be especially helpful. Let me explain. As we said earlier, your usage limits are based on the total length of your conversation combined with the number of messages you send every time you send a message in a chat, Claude needs to reread the entire conversation. The longer your conversation becomes, the more work Claude needs to do to continue the conversation. But here is a trick. If you have a long conversation with multiple artifacts created and want to continue modifying an artifact, or part of it. Instead of doing this in that exact launch chat, you can download the artifact to your local drive and then begin a new conversation with that file as your starting point. This improves Claude's performance by giving it direct access to just the relevant artifact you want to work with without requiring it to process the entire previous conversation context. By the way, Claude will also respond faster in shorter conversations. Let's see how it works. Here is the Melbourne Schools analysis we worked on in the previous lecture. This content was created as an artifact, meaning I can download it as a separate file and refer to it later in any of my future conversations with Claude to download the file, click on download to file. The file will be saved to your local drive. Now, let's say you want to return to working on that school's analysis. What you can do is to start a new conversation Upload the MD file with the previous analysis and specify what changes or refinements you'd like to make to that artifact. For example, this is my request. I'd like to rank the schools based on the following criteria, schools that are known for their achievements in math and science. I also added information that top performance schools should go first. Let's click on Enter. Claude starts working with the original artifact directly, making the requested modifications while maintaining all the original structure and functionality. I hope you love this little trick for working with artifacts and start implementing it for your work. And while we are here on the topic of performance improvement, let me also give you two additional recommendations that work whether or not you are using artifacts. Ask multiple questions at once, especially when working with a lone conversation. Since Claude has to read the conversation each time you send a new message, asking questions in individual messages uses up your limit faster than a single message containing multiple questions. Avoid reloading the same file multiple times to the same conversation. Claude already sees the entire context from your conversation. You only need to re upload a file when starting a new conversation with Claude. Okay, that's it for this lecture. Let's recap the key points. And 19. Follow-along: Creating visuals with Claude – process flow diagram (part 1): God. Good. For this demo, I prepared a description for a fictional company called Mosaic Mind, which is building an AI powered collaborative storytelling platform. The company has recently raised Series A funding and is planning to actively hire new team members. So they require a clear procedure for onboarding new employees. What truly sets the company apart is its remote first culture. Currently, they already have a 30 person team spending 12 countries and five time zones. They don't rely on the traditional nine to five workday favoring a so called hours overlap model. Teams need to be available for synchronous work for 4 hours only with the rest of the schedule fully flexible. Frankly, I would love to work in that kind of company. Let's look at what process Cloude will suggest for this remote first organization. So for this, let me open Cloude and begin the conversation. The first thing we do here is we attach the file with the company description. You have several options. You can choose a file from your local drive or you can make a screenshot, or you can select a file from your Google Drive. My company description is on my local drive. I will just dragon drop the PDF to the chat. Let's also use the newest Clote 3.5 Sanet model, which is the most intelligent version to date. In terms of the prompt, I don't want to give detailed instructions just yet. I'd like to see what clot suggests first and then modify the diagram. I will ask Please create the new employee onboarding process flow for mosaic mind startup, and let's press Enter. Clote starts creating an artifact in the form of a mermaid diagram. A Mermaid diagram uses Markdown style syntax to create various types of diagrams using text. The key advantage of mermaid is that you can create complex diagrams without needing specialized diagram and software. Just write the description in text and you get a visual diagram like we see here. Many platforms like Notion or the mermaid life editor support mermaid diagrams. The beauty of creating mermaid diagrams with Cloude is that it can generate the entire text description for us, and all diagram modifications can be made through text instructions. Notice that Claude also provide a description of the key components of this onboarding process, which is really helpful. We see that the process consists of five steps, including pre onboarding phase, first day focus, core integration elements, step, even though this is not probably a separate step of the process, but let me check this later. Next, we have first week milestone, followed by first month's goals. Claude also suggests what substeps each step could include Claude picked up nicely on the remote first company culture. So you see steps like shipping home office equipment, meeting with other colleagues through virtual tools like Office pot. And we even have step completing training related to the asynchronous work culture. Of course, not all of these steps will be relevant for us, and remember that we did not provide Claude with any specific expectations or requirements for the boarding process. You can add new steps, remove existing ones, change the order of steps, add or modify decision points, change the text of any step, or adjust colors and styles. So you have a lot of flexibility to customize the diagram. Let's make some modifications to the process. The first change that I'm going to do is I'd like to reorganize the steps in the pre onboarding phase. So I'll write the following. So I asked Claude to reorganize the pre onboarding phase, and I included the correct substeps that I want Claude to include instead of the current one. Let's see if Claude makes the job right. Let's hit Enter. And we see that Claude starts creating the second version of the artifact. Let me enlarge the diagram by pressing on the plus sign. The issue here is that I don't want the steps to be sequential, so I want everything happen in parallel. Let me ask Claude to change this. Yes, that's exactly what I wanted to have. Now we have a pre onboarding phase followed by three steps happening in parallel. Let me also make some other changes to the first day of onboarding. All right, so what I asked Claude to do here is to move the assign on boarding body under the virtual office in introduction. And I also want to change the name for this phase. First month's milestone replaced with first month's deep integration. Let's hit Enter and see how Claude will reflect those changes in the process flow. All right. Let's do one more change to the process flow. Here, I've included quite a lot of changes. I first asked for a decision point and then explained what steps should follow depending on whether the review is positive or negative. I also asked Claude to change the color of the key components of the diagram. Let's see if it incorporates all the requests. Notice that I'm using simple conversational language, like I would do with a colleague or assistant if they were showing this process flow to me, and I wanted them to make those modifications. As always, I'll press Enter and let's look what happened with the process flow. Okay. As for the color schema, it's definitely not what I wanted, so I need to clarify my instructions. But let's check what happened with the decision block. Yeah, this loop seems to be correct. So now let me make modifications to the color schema. All right. After making a lot of changes to the color schema of this process flow, I think we got great results. We see that Cloud incorporated the changes that we requested. Every time we make a modification, Cloude creates a new version of the artifact. We can scroll through the versions to review changes made in previous iterations and continue modifying that version by typing in our instructions into the chat. What I recommend is grouping several modification requests into one message instead of sending them one by one, even though we didn't see this in this specific demo but based on my previous tests, I know that Claude y sometimes redraw steps incorrectly by mistake, even without asking to modify those steps. So to avoid such behavior, try to send your requests in bulk. All right, this tutorial is becoming quite lengthy. So let's take a break here and meet again in the second part of the tutorial. I'll see you there. 20. Follow-along: Creating visuals with Claude – process flow diagram (part 2): Everyone. Welcome back to the second part of the tutorial where we work on a process flow for a new employee onboarding process for a startup called Masaic mind. If you missed the first part of this tutorial, please watch it first before starting this video. After making a few modifications to the process flow, our chat can become quite lengthy. So as we discussed earlier, to improve clouds performance, you can download an artifact to your local drive or save it to your Cloud storage, and then continue modifying it in a new chat. Before downloading an artifact, make sure that the correct version is selected. For this demo, I'll choose the last version and click on Download to File. Now let's say that I want to make modifications to that artifact. I'll start a new chat. Attach my artifact to the input data for this conversation, and then type in my instructions. I don't remember the exact name of the step where I want to replace notion for confluence. But let's see if Claude will be able to pick it up. I and here are the changes. We see that the process flow has been updated by replacing notion to confidence in the grand access to documentation software. It's great that Claude maintained all other elements of the onboarding flow without changes. Apart from making modifications to the process flow, you can ask Claude to create more detailed guides or templates for any part of the process. For instance, I can ask the following. Please remember that since this is a new chat, Claude does not have knowledge about Mosaic mind, which we provided in the form of EPDFfle in the first chat. You can avoid reloading the documents by using Claude project functionality, which we'll cover very soon. But for now, let's re upload all the files Claude will need to design the template. Our document is here in the chat, and I can press Enter. And here we go. Here is a template we can use as this, or we can make changes to align it with our workflow. Notice that Claude has created the second artifact for the same chat. And in case if you have several artifacts created in your conversation, you can easily choose the one you'd like to work with by clicking on chat controls. And from here, choose an artifact you'd like to modify. All right, our tutorial won't be complete unless we cover how to quickly export the diagram of the process flow to include it in the relevant documentation. I'll show you how to export it in Notion, a collaboration tool for note taking, knowledge management, and data management, as well as project and task management. However, if you don't use notion in your workflow, you're welcome to use other tools, including mermaide Live editor or any text editor with mermaid support. Here is a new document I created in Notion for this demo. Begin by clicking slash and from here, select code. Let me just type it. And in this window, choose mermaid. It's already selected for me. Now come back to Clode. Find your artifact with the flow diagram. Click on code and then click Copy Content. Next, return to notion and paste the code in this window. Let's click on Review. And now we have our diagram exported in Notion. How cool is this? You can work on this document by including any text to your diagram. Let's type something here. In case you want to modify the diagram again, you can always return to Cloud and continue the editing process through the chat. In case you are not using notion, you can easily export the diagram as an image in JPEq or PNG format. For this, open the mermaid live editor and replace the default code with the one generated in Cloude you will see the visual representation of the process flow on the right hand side of the screen All Good here to download the file, go to Actions and choose the format you want the diagram to be saved in. Let me choose an SVG from here, and we see that the diagram has been saved to my local drive. Now we can open a file. And continue it in any software where we prefer to keep our project documentation. All right. I hope you enjoy working on creating visuals with Claude and explore this functionality for your work processes. I would recommend doing some practice right after this tutorial, as trying things out right away is the best way to get the hang of it, and I'll see you in the next video. 21. Publishing and remixing artifacts: One and welcome back. Let's continue learning about clothes artifacts. Apart from creating and modifying artifacts, you can also make artifacts available to other clothe users and you can remix artifacts created by others. Let's see how it works. Publishing an artifact allows others to view and remix your work. To publish an artifact, find the one you want to publish and if that artifact has several versions, find a version that you want to make public. Next, click on the published button at the bottom right side of the artifact. Let's click on Publish and copy link here. And now you have a link that you can share with your friends or colleagues so that they can work with your artifact as well. Of course, before you publish an artifact, double check all the content included to ensure that it doesn't contain any private or confidential information. Please note that only the artifact itself is published without the surrounding conversation and other context from your chat. If you realize you made a mistake and don't want the artifact to be publicly available, you can always unpublish the artifact. To do this, click on published first and on the pop up screen that will appear, click on unpublished button. However, you won't be able to republish that artifact again, you'll have to create a new one now let's talk about remixing artifacts. You can remix only published artifacts. Closed documentation states that all published artifacts will be viewable on a separate public facing website. However, at the time of recording this tutorial, this website is not available yet. How do you find the published artifacts that you can remix? Well, several third party websites have emerged with collections of such artifacts. Here is a website with a collection of code based artifacts. Another one presents artifacts from various categories, not only in programming, but also creative fields, lifestyle, education, gaming, and others. Let's use this platform to browse through the published artifacts and remix something we find useful to start, go to clodartifacts.com. From here, select a category. For my example, I would like to remix a game. Browse through the list of games. Here I see Tetris. Let's view this artifact and think if you would like to remix it. Wow, this is exact the same variation of tetris I played for the very first time. It's so nostalgic to play it after so many years. I feel like I could play it endless lim. Let's definitely remix this game by pressing the remix artifact button. Claude starts remixing the artifact. Yes, all good and we see that the game has been remixed perfectly. We also see that Claude gives us some ideas for modifications we can make to the original game, which is super helpful. Let's ask Claude to implement two standard tetris features. Next piece Q and hold piece. This would add strategic depth and planning to the gameplay. I'll add the following instructions. So I asked Claude to add next piece Q and hold piece. I also explained how those features supposed to work. It is so scary to see how the blocks randomly fall down, but I'll try to ignore this for now and fully concentrate on my conversation with Claude. Let's hit Enter. We see that Claude starts creating the third artifact to incorporate the changes we just requested. Go. Right, so the wholet piece seemed to be working very nice, but I don't see the queue of three blocks over here. So let's ask Claude to fix it. All right, I like how our game is shaping up. You can also review the suggestions from Claude in terms of the features that can be included into the game and ask Claude to add any of these. Now, in case you want to share the game, you can publish the artifact and then share its link with your colleagues or friends. If you want this game to be available on a separate website, you may need some programming knowledge to export the game from Clote. I tried several no code options for exporting the game that Clote suggested, but I didn't find any easier solution for this. All the options were quite convoluted and still required some technical knowledge. I'll update the tutorial as soon as I find a straightforward no code solution. All right, that's it for this video. Now it's your turn to get to work. What I suggest is that you browse through the clode artifacts, showcase website, and remix one or two artifacts. Also, look through the artifacts you already have and decide if you are ready to publish something so that other users can see and remix your creation. You can go even further and submit your artifact to the artifacts showcase website by clicking on the submit Artifact button. I hope you enjoy this practical work, and I'll see you in the next video. 22. Projects: Definition and how Claude uses them: Hello, everyone. Now that you know what artifacts are and how to create, modify, publish and remix them, let's learn about another useful feature of Claude projects, which can help you organize your work with Claude. Before we start, let me warn you that projects are available in the paid plans only. So if you are currently using Clouds free plan, this is not something you can experience just yet. I would advise you to skip the upcoming lecture, focusing on projects functionality, and revisit it later if you choose to upgrade to the paid subscription. With that set, let's get started. What exactly is a project in Clote? Think of a project as a dedicated workspace or folder where you can organize related conversations with Claude, just like how you might create different folders on your computer for various clients or tasks. Projects help you keep your AI interactions neatly organized. Creating a project is simple. When you are in clouds interface, look for the project icon in the left hand sidebar, click it. Give your project a name and description, and you are ready to go. For example, if you are working on a marketing campaign, you might create a project called Marketing Campaign to keep all related conversations in one place. Now, let's figure out the benefits of creating a project. One key benefit is project knowledge. The ability to provide context for your chats with Cloud. You can upload relevant documents, text, code or other files to a project's knowledge base, which clot will use to better understand the context and a background for your individual chats within that project. In addition to project knowledge, you can define project instructions for each project to further tailor Cloud's responses. For example, you might instruct clade to use a more formal tone or answer questions from the perspective of a specific role or industry. These project instructions will work alongside user preferences we said earlier in the profile settings, as well as the selected style in a chat. The third useful aspect of projects is context sharing. When you're working within a project, Claude remembers important information from previous conversations in that project. For instance, imagine you are developing a new product feature. In your first conversation, you describe the feature requirements and in your next conversation within the same project, Claude already knows these requirements, so you don't have to repeat them. Projects also make it easier to collaborate and keep track of your work. Each project maintains its own history of conversations, which you can easily reference later. This is particularly useful when you are working on long term tasks or needs to revisit previous discussions. When does it make sense to create projects instead of standalone chats? Basically, you can create a project to organize any tasks, both work or personal that can benefit from shared context. At the same time, avoid creating projects for one of tasks where using an individual chat will be more efficient. For instance, if you are a marketing manager, you might have one project for blog posts, another for social content, and a third one for email newsletters. This separation helps you stay focused and makes it easier to find specific conversations later. And if you are in educational content creation like me, you could have one project for research and themes and topics for your next course based on research papers, industry reports, and other available information. You could also have a project for planning and drafting the educational content, including lesson plans, lecture notes, exercise problem sets, and quiz questions. Projects can also be helpful for organizing personal tasks. For instance, as mentioned earlier, while recording this course, I was planning a relocation to Melbourne, Australia. So I set up a dedicated project in Claude to help me plan the entire move from deciding which residential area best suits my family's lifestyle and selecting a school for my daughter to create enough to do list for tasks after arriving at the new location. Now think about which parts of your tasks can be organized via projects. And while you are thinking about the projects use cases, let's sum up this lecture. 23. Expanding your prompt engineering skills even further: Section Intro: Welcome to this section on Advanced Frampton Techniques in Cloud AI. We are going to kick things off with step by step thinking, also known as chain of thought Frampton. This approach helps you guide Cloude through complex tasks in a structured and logical way. Then we'll cover a very important topic. Minimizing AI hallucinations, while Cloude is powerful, it's not perfect, and I'll share practical strategies to keep its responses grounded and reliable. Be hands on practice is key, we'll have a follow along exercise where you will learn how to summarize long form content with clade step by step. By the end of this section, you will feel even more confident in crafting prompts that get accurate actionable results. Let's dive in 24. Thinking step by step or chain-of-thought (CoT) prompting (part 1): One and welcome to the new lecture. Let's explore another powerful prompting technique for working with EI assistants like Claude. The thing, step by step approach, also called chain of thought prompting. Imagine you're teaching a child how to solve a complex math problem. Would you just want their final answer or would you want to see their work? Just like watching students thought process helps us understand their reasoning. Chain of thought prompting helps us get better more reliable results from AI models. In this video, we will cover what chain of thought prompting is, why it matters, and how to implement it effectively. Let's get started. So what exactly is chain of thought prompting? At its simplest, it's asking AI model to explain its reasoning step by step rather than just giving you the final answer. It's like having a conversation with an expert consultant. You don't just want their conclusion. You want to understand how they reached it. Research has consistently shown that when AI models explain their reasoning, they perform better at complex tasks. But why? Well, have you ever noticed how explaining something to someone else helps you understand it better yourself? Same with the AI model. By asking the model to break down its thinking process, we give it more space to work through complex problems. Now let's look at how to implement chain of thought prompting. The easiest thing you can do is to add a thing step by step request at the end of your prompt. This is so called basic chain of thought prompting that is quick to implement and that works for simple tasks. However, the downside of this basic technique is that you don't tell Claude how to think, which is especially useful if a task is very specific to a certain use case, business process or organization. So the best practice will be instead of just saying things step by step to outline the actual steps you want the AI model to take or think through. This technique is known as structured chain of thought prompting. By asking a model to generate a chain of thought or a series of intermediate reasoning steps, we can significantly improve the model's ability to perform complex reasoning. Let me share an example that illustrates chain of thought prompting. Imagine you are asking AI to help you choose between two business strategies. Here is how the same question could get very different results. Here is a basic prompt. The model might simply respond with a preference, but you won't know how it reached that conclusion. Now let's look at a chain of thought prompt. This structured approach forces the model to show its work, making its recommendation more valuable and trustworthy. A Please note that chain of thought prompting is not just about getting better answers. It's about getting more reliable and verifiable answers. When you can see the model's reasoning, you can spot potential errors or biases more easily. All right. And that's it for this introductory lecture on chain of thought promptin. In our next lecture, we will discuss several advanced techniques for chain of thought promptin. But before that, let's sum up this lecture. 25. Thinking step by step or chain-of-thought (CoT) prompting (part 2): One and welcome to our second lecture on chain of thought prompting. As promised, let's talk about more advanced techniques for chain of thought prompting. One powerful approach is to combine it with role playing, which we already covered earlier in the course. For instance, you might ask the model to think through a problem as different stakeholders. This multi perspective thinking often reveals insights that a single chain of thought might miss Now, let's talk about another technique, fuse shot chain of thought prompting. In this approach, we include several worked examples of how to solve similar problems step by step within the prompt. These examples act as a guide, teaching the model how to reason through the task. Let's look at an example. Notice that problems in examples one and two are different as a few shot chain of thought prompting isn't about showing examples that are identical, but about teaching a generalizable thinking framework. The examples demonstrate how to think, not just what to do. Even if the context changes, the model learns to break down the task logically, assess relevant factors and reach a reasoned conclusion. By seeing this pattern in multiple context, the model can generalize and apply the framework to new unseen problems. Similarly to few shot prompting, there is one shot chain of thought prompting where you provide just a single example to guide the model's reasoning. This technique may come in handy when your task is straightforward, but still benefits from structured reasoning. And finally, before we finish this lecture, let's explore some common pitfalls people make with chain of thought prompting and how to avoid them. Pitfall number one, not verifying the reasoning. Remember, just because the model shows its work does not automatically make its conclusions correct. Always review the logic in each step with fall number two, overcomplicating the structure. While structure is important, too many steps can actually confuse the model. Aim for about three to five main steps in your thinking process. All right. That's it for the chain of thought prompting technique. So next time you ask AI model a complex question, don't just ask for the answer. Ask it to think step by step and show its reasoning. Pay attention to how this changes the quality and reliability of the responses you receive. Finally, as always, let's sum up the key points of this lecture. 26. Keeping it real: Practical strategies to minimize AI hallucinations: One. Imagine asking an AI assistant about the recent news event and it confidently cites a detailed article that does not actually exist or asking it about public figures and getting responses that mix real facts with completely made up details. These aren't bugs or glitches. They are what we call hallucinations in AI, and they are one of the biggest challenges when working with language models like clause let's explore why these hallucinations happen, how to spot them and most importantly, practical techniques you can use right away to get more accurate, reliable responses from Cloud. Let's start by understanding why AI models sometimes hallucinate. Here is what happened. Language models are trained to recognize and complete patterns in text. Sometimes they will extend these patterns in ways that seem logical but aren't factual. Models like Claude are trained to be helpful and to provide complete answers. Sometimes this helpfulness instinct overrides the ability to say, I don't know. While Claude has been trained on vast amounts of data, it has a knowledge cutoff date and cannot access real time information. When asked about topics beyond its knowledge, it might try to extrapolate based on what it does know. Now that we understand why hallucinations happen, let's explore how to spot them in practice. Think of this as developing your AI fact checking skills. Once you know the warning signs, they become much easier to catch. Here are the key warning signs to watch for. Overly specific details. When clot provides very specific details, especially about recent events or statistics, this should trigger extra scrutiny. For example, if it gives exact numbers for market data from after its knowledge cutoff date, that is a red flag. Perfect sounding citations, examples or statistics. If you notice an answer that sounds too perfect, that's a good reason to double check the information. And believe me, the more experience you become working with clothes and similar AI tools, the better you will be at spotting these two good to be true moments. You will develop an instinct for recognizing when something feels off or overly polished and that's your cue to dig deeper, verify facts, or cross check sources, trust but verify. That's the golden rule when working with AI generated content, inconsistent answers. If you ask the same question multiple times and get different specific details each time, that is a strong indicator of hallucination. Overly definitive statements. When Claude makes very definitive statements about topics that should have some uncertainty, especially regarding future events or complex topics, be cautious. Knowing why hallucinations happen and how to spot them is a great start. But how do we actively prevent them? Let's go over four useful strategies that will help you get more reliable, accurate responses from Claude every time. Strategy number one, be explicit about uncertainty. For instance, instead of writing, what were the key findings of the Johnson's report from 2024? Try this. If you are familiar with the Johnson report from 2024, please share its key findings. If you are not certain about any details, please let me know. Or instead of list all the companies using this technology, try this. Based on your knowledge cutoff date, can you list some verified examples of companies that we are using this technology? Please indicate if you are uncertain about any examples. Instead of what's the market size for AIhatbard in 2025? Try this. Can you provide market size estimates for AI chat boards from reliable sources within your knowledge cutoff date? Please specify the time period for any numbers you share. Notice how each revised prompt explicitly gives clod permission to acknowledge uncertainty and limitations. This simple change can dramatically improve the reliability of responses. Strategy two, request citations and reasoning. When working with documents or data, ask Claude to site specific sections or explain its reasoning. This is like asking your research assistant to show their work. It helps you verify the information and catch potential hallucinations. Let's look at the example. As you analyze this document, please quote specific sections that support your conclusions. If you make any interpretations or extrapolations, explicitly label them as such. Strategy three, use structured output formats. Requesting structured outputs can help minimize hallucinations by forcing Clote to organize information more systematically. For instance, please analyze these sales data using the following structure, verified data points, direct numbers from the document. Calculated metrics show your calculations, interpretations, clearly labeled as interpretations, uncertainties, areas where data is unclear or missing. Strategy four, implement verification steps. Include verification steps directly into your prompts to enhance the accuracy and reliability of clouds responses. For example, you can ask clots to list any assumptions it made during its analysis. Highlight areas where it has lower confidence or certain them. Recommend additional information that could help validate its conclusions. This approach ensures a more thorough and transparent output, making it easier to assess the quality of the responses. All right, now that you have all the information on AI hallucinations, take a moment to review one of your recent prompts. How could you modify it using the strategies we've just covered? And remember, the goal is not to eliminate hallucinations completely, but to create a workflow where they are less likely to impact your results. Please share your original and revised prompt under the Q&A section for this video. And as always, let's briefly recap the key points of this lecture. 27. Follow-along: Summarizing long-form content: One. Welcome to our new tutorial. Here we are going to work on summarizing loan form content with Claude. Here is what I'd like to do. I have an interview with Dario Amade CEO of anthropic, the company behind Claude, where he talks about many topics, including artificial general intelligence and the future of AI and humanity. The interview is quite lengthy, it may be challenging to find the time to listen to everything thoughtfully. So I'm going to ask Claude to prepare a detailed summary analysis of the interview, including strategic takeaways on industry trends identified and leadership insights. I also want Claude to prepare a reference section with all books, articles, and resources mentioned in the interview. I think this task is perfect for a chain of thought prompting, as it's quite complex and requires multiple steps of analysis. And since Claude does not transcribe videos, I have copied this video transcript. For ex interviews, he usually includes video transcripts on his website. So I copy them to a separate file. I'm going to use this file in my conversation with Cloude. Alternatively, you can copy paste the script for a YouTube video directly by first clicking on view four chapters. Next, you will see a video transcript, which you can copy to Cloude. All right, we are all set with the preparation. Let's start by asking Claude to divide the transcript into sections or topics. Let me also attach my transcript one more time since I just started a new chat. All good. And I press Enter. Large unstructured content can overwhelm the AI model, leading to scattered or incomplete results. So by identifying the main sections, Claude creates a clear roadmap for the analysis, ensuring that no key areas are overlooked. We see that Claude has divided the transcript into ten different sections, providing time codes for each of the section and it also included brief information on the key topics discussed in every section. Now, let's focus on one section at a time to extract detail insights. Let's begin with the section one. Focusing on one piece of content at a time provides depth and clarity for each part of the transcript. By isolating individual sections, Claude can dive into specifics without being distracted by the rest of the content. Here we see quite detailed summary of the first section. And let's do the same exercise for the remaining nine sections that Claude identified in its previous response. After analyzing all sections, let's ask Claude to synthesize insights into actionable themes. Combining data from individual sections helps generate a cohesive summary of actionable takeaways. This step ensures the final output reflects broader trends and over arching lessons. Please also notice that each chain of thought prompt builds on the previous information from our conversation with Claude, right? Here are the strategic takeaways from the interview. Here we see core principles, emerging industry trends, followed by leadership insights, product development methodologies, and team collaboration approaches. Now let's focus on resources and references. This step will make the output more comprehensive and useful for follow up. All right, we've got a list of resources mentioned in the interview. However, since the cutoff date for the newest 3.5 sanat model is April 2024, Claude is saying that it cannot guarantee that all the links will remain active, so we have to double check them. Write grade. And as an optional final step, we can ask Claude to produce a polished final summary. So from this demo, we've seen how the structured approach of chain of thought prompting enhances the AI's reasoning capabilities and output quality. I hope you start implementing this technique in your daily work. As practice is the best way to retain information, please go ahead and summarize a video or another piece of loan content just after you finish this lecture and that's it for this demo, and I'll see you in the next one.