Transcription
While ChatGPT competes with other neural networks, CLOT is quietly but confidently growing. Artificial intelligence, created by former Open AI employees and funded by Google, a brother and sister assembled a team of top Open AI engineers and released Claude with Constitutional AI. This means that if ChatGPT simply tries to please people, but doesn't always understand why something is right or wrong, then Claude, before giving an answer, undergoes special filtering, that is, through a constitution, which has more than fifty clear principles. And because of this, the answer becomes much higher quality than in ChatGPT. But is everything ideal in Claude, in what is it better than ChatGPT, and in what is it worse? I analyzed all the life hacks, and in this video, I will tell you only what is truly useful for everyone, not just programmers. We will figure out which models you need for your task, how to create your own assistants within Claude, how artifacts help you create applications, documents, games, and generally simplify your communication. I will also share secret life hacks that allow you to use Claude 100%. At the end of the video, I will tell you why you should not use Claude for all tasks. All the information from this video, as always, will be on my website. After you log in to the site, the familiar chat window opens. But there are a couple of distinguishing features here. Firstly, we can switch between Claude's Sonnet, Opus, and Haiku models. For Sonnet, we see an inscription. It is ideal for everyday tasks, and it is set by default. This is indeed true, because this model provides a balance between speed, quality, and cost of integration. In general, this is the only model I use for everyday tasks, for analysis, and for working with files. Opus is officially recognized as the best for programming, but it is also suitable for more advanced analysis. The third model is Haiku. It has lower quality compared to other models, but it is the fastest and cheapest, so it is often used in chatbots for stores, for moderating comments, for quick fixes of code or text without deep analysis. The cost of using models for integration is also available on the website. Claude has a larger context window than many versions of ChatGPT. What does this mean? A context window is the size of the dialogue within your chat with artificial intelligence, during which all the information you have written within one dialogue is remembered. In ChatGPT, you have likely encountered a situation where you write it a task, taking into account the previous context and messages, but it writes you general information in response without reference to your context, or vice versa, it writes what you don't need. And the whole point is that you have reached the context window limit, and the previous part of the context in the dialogue has simply been deleted, and a new context window has opened. And you waste time to re-explain the context and clarify the task to it again. So, Claude remembers information better during the dialogue. And if we exceed the context window limit, it simply pauses the chat and does not allow writing further in this dialogue. By doing so, Claude protects us from stress and wasted time on re-explaining the context. Interesting life hack. If you believe that you have received enough valuable information within one dialogue, then I recommend creating a document with key ideas and context, simply by asking Claude to do it. This text document can be used in new chats or simply saved in your note-taking program, for example, in Obsidian. Documents that Claude creates are called artifacts. By the way, working with artifacts in Claude is implemented much cooler than working with ChatGPT's Canvas. Firstly, as a screenwriter or student, I often encounter the fact that ChatGPT does not accurately determine the number of words or characters in messages. For example, I often ask it to reduce the size of my text to 3,000 words, but instead I get a text snippet of 700 words. But I haven't seen such a problem in Claude. It creates and formats the text for me in the form of an artifact. And all these artifacts are stored in a separate place. On the left panel, we see my artifacts. Here you can store the key ideas of all dialogues. Secondly, with the help of artifacts, you can create interactive games and applications. And this works much cooler than we did it in ChatGPT, because Claude writes code much better. For example, let's ask it to create a game based on a book on self-esteem. If we try to upload the book directly into the chat, it may give an error that the file exceeds the allowed size. For this, we simply upload this file to our Google Drive and open this file using documents. Then we open access to it and share the link to this file in the chat with Claude and write him this prompt, which will also be on my website. And it creates an interactive game that will help us not only remember the material read in the book, but also gradually learn to implement the principles in life. Thirdly, artifacts perfectly handle documents. It can create checks, receipts, which are very realistic, but, naturally, as samples due to the principle of law-abidingness. However, if you need to translate a document into another language, while preserving all the structure, data, and formatting, then it will cope with this perfectly. It also makes minimalist presentations with animation, charts based on your data, for example, it will analyze your financial situation and give you recommendations. I also like that when a large answer is planned, it transfers it to a separate artifact so that it is easier for me to read it. This artifact can not only be downloaded as a text document, but also shared with a friend via a link. Claude accepts text documents, photos, and can connect to third-party services. Text files, for example, books and PDFs, it may not accept if they exceed a certain size, for example, 100 pages. And if it exceeds the size, then you can upload it to Google Drive, as we did before, or upload projects. More on them later. Claude can also accept and analyze information from email, from Google Drive, from the calendar. Just like Gemini does, for example. I sometimes ask it to summarize the last email in my inbox and save it to artifacts. Recently, connectors have been added to Claude, that is, tools for connecting Claude in one click to popular services. You can connect Notion, Figma, Canva, and other tools and use them as no-code tools for creating designs, texts, which will immediately appear in the corresponding platforms. By the way, about no-code. We recently had a video on the channel where we explained what it is and how you can automate your life with its help. If you are interested in how Claude helps automate your life with the help of third-party services, write about it in the comments. I will definitely record a video on this topic. Claude can be configured in the style of response depending on your mood and desire. If in ChatGPT we simply indicated the general style in the settings, and it adhered to this style in all dialogues. Even when I needed to simply fix a door handle at home, it would give me motivational phrases for no reason. So, in Claude, you can set the response style situationally. For example, for responding to an email in English, create a specific style. And for writing video scripts, also use a separate writing style. Many users create their own style to maintain a consistent brand voice when creating advertising texts, social media posts. To create styles, click the tools button, select use styles, and here by default are the styles that I also use: Normal, Concise, Formal, and Explanatory. Normal is standard answers, nothing more, nothing less. Concise is short answers, strictly to the point. Formal is more scientific and formal answers. Explanatory is my favorite educational answers, which are well suited for learning new information. Here we create our own style. Click the create and edit styles button. Here we can simply upload a pre-prepared instruction on how Claude should respond to us, or simply paste a file with the text of your favorite blogger or other source. And Claude will also respond as in this file. Or we can create response styles using Claude's prompts. I will choose the last prompt and write that it will be a psychological style. And I want to know the name of the psychological problem, the ways to solve it. And let it ask five clarifying questions. Click create style. Here I change the style name to my own. Then I save it and use it in subsequent chats. Let's send him a message that we often lie down instead of taking real action. And we see that he responds like our personal psychologist. Such prompts, by the way, are already on my website in the prompts section. You can take them from there and improve them for yourself. Claude also has an advanced thinking function. It is enabled by default in all my requests. It allows me to see Claude's reasoning logic for solving my problem. It combines well with research mode and the explanatory style. With these three settings, research is thorough, with up-to-date information, and with clear presentation. Unlike ChatGPT O3, where the thinking process is shown quite vaguely, Claude breaks down each step, and we see exactly how it arrived at this conclusion. Projects in Claude, in my opinion, work much more effectively than in ChatGPT. It is ideal for repetitive tasks. For example, I have a YouTube channel, psychological analysis, website optimization, and other projects. Here you set a system instruction for the specified project. Add files, books, texts, attach files that are on your Google Drive or GitHub. I like that you can access projects from a regular chat. You don't need to go anywhere. Just click the plus icon and select the corresponding project and write your request. Interesting life hack. You can automate the input of large prompts using applications like TextExpander or Raycast. You type one word, and a long prompt is inserted. To do this, you need to configure it in the TextExpander application, enter the full text and write a replacement word. This is especially useful when routine is repeated within a project. For example, conduct research on this topic. The only serious problem is the limitation on the free plan. On the free plan, you can write four to five requests per day. In general, you can experiment little on the free plan. However, the cost of a paid subscription is the same as for ChatGPT, $20 per month. But even paid subscriptions have a limit on the number of requests. The Max subscription removes almost all restrictions, but it is very expensive, $200 per month. Therefore, I use a combination of Claude and ChatGPT with the help of artifacts that contain the key ideas of the dialogue. Claude, like other neural networks, is not yet ideal and universal. Despite the large context, it can still forget dialogue details and in general sometimes even gets confused in short dialogues. Therefore, artifacts with key dialogue ideas save the day. I cannot recommend it for all tasks, but it excels at deep analysis, programming, and education. And in general, with the help of writing styles, you can create personal assistants directly within Claude. Want a psychologist, want a lawyer, or want an analyst. Personally, I started using it for learning programming, because the games and material it creates are more informative than what ChatGPT does. Working with files is convenient. Especially if you connect Google, Notion, Figma, and other no-code tools. Thus, I use it for advanced analysis, for learning, and for writing long texts. GPT for quick and everyday tasks. Gemini for interaction with the Google ecosystem and creating videos. Subscribe to the channel. In future videos, we will look at the interaction of Claude with other services.