Transcription
Hello everyone. Today we have a serious conversation about efficiency. You know, I've noticed a dangerous trend. We've all fallen into the trap of instrumental thinking. Every week a new neural network or update comes out. And we think: "This is it. If I buy a subscription to this service, my productivity will skyrocket, my business will boom, and I'll work 2 hours a day. Or artificial intelligence will do everything for me. We're buying subscriptions. CHGPT, Cloud, G Mini, Perplexity. How many of you have more than two services you don't really use? Write about it in the comments too. I'll tell you right away, the problem isn't a lack of tools, there are hundreds of them. The problem is the absence of systemic thinking. More neural networks does not equal more productivity. Often it's just more chaos and less money in your account. I've been in the artificial intelligence field since 2022, even before it became mainstream. And as a programmer, I don't look at pretty interfaces and marketing, but at how these models work under the hood. I've tried hundreds of tools not for hype, but to find those few that actually integrate into the workflow, not just drain money. Today I will show you not just a set of services, I will show you specific, proven, working workflows that I use every day. We will cover first, how to conduct deep market research, not get superficial answers. Second, how to visualize strategies. Third, how to create ready-made assets from presentations and reports to videos on a topic in minutes, not weeks. Before we start diving into practice, do one simple thing. Subscribe to this channel. I'm not here to hype on news, but to give you tools that make money and save time. Clicked the subscribe button. Great. Then let's begin. The first thing you need to do is stop asking which artificial intelligence is the best. This is like asking: who is better: a surgeon or an architect? It depends on what you need: to remove an appendix or build a house. I suggest you treat neural networks as hired employees. Each has its own specialization. Here's my team. Chat. This is a universal specialist. He knows a little bit of everything, can write texts, code, but can sometimes be superficial. Gmi from Google is an integrator. It has a huge memory, a large context window. You can feed it an entire book or a huge report, and it won't get confused. Cloud is a strategist. If you need beautiful text, complex analysis, or code that looks aesthetic and works, go to him. Plus, he works best with language nuances. Perplexity is a researcher. It doesn't invent, it searches. It has access to the real internet right now. Notebook LM is a data analyst. It works strictly with the documents you give it. Nothing from itself. And here the question arises: "Ivan, this is all cool, but how to pay for all this from Russia? Cards don't work, accounts get blocked, dollar prices bite." Friends, I've solved this problem for you. In the description under this video, I've left links to trusted sellers. Through them, you can purchase subscriptions to any of these services. CH GPT Plus, Cloud Pro, Perplexity Pro, GMI Pro. So, if you like any of the tools from the video, you know where to get them safely and quickly. Link in the description. And now let's get down to business. I'll warn you right away, you'll definitely need a VPN turned on for all these services to work. Friends, sorry, but these are the harsh realities. Without it, access to technology is currently closed. Case 1. Let's start building a system. Most people just write to the chat and get garbage. We do it differently. Go to the chpt projects section and create a project. Name it furniture brand hardware X. This is the foundation of our context. Upload your knowledge base here. In my case, it's just a strategic analysis, but if you have PDF files, a brand book, and interviews with the founder, upload them too. Now the model doesn't answer abstractly, but strictly based on your files and numbers. We ground the neural network. But let's use not the usual chat here, but the Deep Research function. Give the task: conduct a market analysis of the premium smart furniture brand. I've also left this prompt in my Telegram channel. Link in the description. Go and get it. With deep research, CHGPT goes online, verifies facts, and compares them with our strategy. The research doesn't happen quickly, but the result will be drastically different from the usual chat answers. You get a real report with links to data from authoritative sources that can be implemented in business. And now the main engineering nuance. We don't stop. We download this generated report and upload it back into the same project. We close the learning loop. With each new document, the system becomes smarter and understands the topic more deeply. Now, based on the enriched data, let's create another document. And I'll enable the host function. Now ask: create a portrait of a B2B client. We get a ready-made document. The artificial intelligence knows about our pain points, and the key insight here is that the client buys not tables and chairs, but feelings and prestige. We can download this file again and add it to the project, strengthening the knowledge base for future tasks. Final test. Task T voice Guide. The system sees past reports and writes communication rules specifically for us. This is not a template from the internet, this is a working tool. This is not a miracle. It's just competent work with data and the right workflow. You build a system, not just chat with a bot. In the free version of Charge GPT, there's a light de search. It's not bad, but there are few requests and the model is old. For production and deep analytics, we need the heavy Deep Research. It's only available in plus and pro versions. And it's critically important when you need to build complex chains of reasoning and minimize hallucinations when working with a large context. If your workflow is tied to chat, the upgrade pays for itself in speed and quality. Especially with the release of CHGP version 5.2, the models have finally been perfected. Compared to the rather outdated 5.0 and 5.1, it's already a working tool, not a budget drain. Tests confirm this. I've been working closely with char GPT for 3 years. Tested different models. If you're interested in a detailed technical breakdown, features of charge GPT, and non-obvious tricks, let me know in the comments. I'll definitely record a separate material on this topic. Case 2. Analytics. We start with pure data collection. Chat GPT is useless here. It starts making things up. We need facts and lots of links. Let's go to Perplexity - it's the best AI search engine. Type in a query. Find the top 20 expense management platforms. Output only a list of URLs of their main pages. Look, it read the live internet and gave us 20 relevant platforms. We're not reading this now, we're just copying these links and text. We need a data array. Go to Google Notebook LM, create a new notebook, and paste the copied links as sources. But there are limitations on the free plan, 50 sources, on Pro it's already 300 sources. So watch what you add. What's the trick? We limit the artificial intelligence. Now it only knows what's written on these twenty sites. We physically disable the possibility of hallucination. Data loaded. Now, brain setup. In the system instructions, I set the role. You are a consultant for B2B product growth. This is important so that it analyzes sites not as an ordinary person, but through the prism of business metrics. Ask a complex question. Compare these 20 sites and identify five unique selling propositions. Imagine how much time it would take a person to read twenty landing pages. Notebook LM does it in seconds. And most importantly, look at the footnotes. Every statement is confirmed by a link to the source. Let's go deeper. Do a SWOT analysis. We turn scattered web pages into a structured strategy, but reading text is boring. Go to Studio. This is a new feature. With one click, we generate a mind map to visualize connections. And let's also order an infographic. Need to send a report to the boss. Click reports and choose the format. We get a ready-made text and copy it to Google Docs. Formatting is preserved. There's also an infographic. In one image, the entire essence of the topic you're currently working on will be collected. I must say the quality is good. Nanobona Pro is Google's flagship model. Everything is generated wonderfully, even in Russian, without errors. And the final button is presentation. Notebook LM takes everything it has read and generates slides. Yes, the design is simple, but the structure and facts are already in place. Result: in 3 minutes, we went from Googling to a ready-made presentation, without ever lying. In the next blocks, we'll scale the task. I'll show you how to automate presentation creation in Google Slides, use Notebook LM for learning, and even for assembling an MVP product. The tool allows you to close these tasks very quickly. Under the hood, GMIA 3 Pro is running - this is the very model because of which Openi declared a code red and urgently released GPT 5.2. And although the GPT update turned out to be decent, G Mini is objectively winning now due to its work with a huge context window and multimodality. For us, the competition of giants is a plus. We get powerful tools to work with today. Case 3. Let's move on to data visualization. The task is to find the real pain points of marketing agency clients. If you ask a regular AI, it will give you banalities. Therefore, let's go to Perplexity. And I'll enable the social filter. We're looking for discussions on Reddit and forums. We need the raw nerve, not SEO articles. Look at the output. This is not theory, these are real people's complaints. Lack of transparency, financial dishonesty, poor communication, and so on. This is a ready-made structure for our future content. Turn this search into a set. Export the data to PDF. We need to transfer this context to the next AI without loss. Now let's go to GMI. Why here? Because it has the best engine for working with code and layout. Enable canvas. Upload our PDF report on client pain points. Give the command "Open this report in canvas". And only after it creates the document, we can choose an infographic for this document. Jmeniia analyzes the file and lays out an HTMLSS structure directly in the browser window. Look, it has automatically broken down the data into charts, highlighted, for example, the transparency trap or red flags. This is already a ready-made quick report or even slides for a presentation. But we, engineers, go further. I don't want to take a screenshot that will lose quality. I ask Gemini, write code and add a button to download this canvas as a PNG image. It writes the script on the fly, a button appears, one click, and I have a high-resolution file on my computer, and the entire page is on it. Result: from complaints on Reddit to professional infographics in 3 minutes. Without designers. If you've had an insight now and thought: "I didn't even know this tool could do that," give it a like. For me, this is the main metric that this format is useful to you. And check your subscription so you don't miss the next episodes. There we will move from manual requests to full automation. Case 4. Let's move on to high-level analytics. I need global trends, but not journalists' articles, but primary sources. In Perplexity, I use the search operator File Type PDF. This is a command that filters out garbage and searches only for PDF files. Look, we immediately get McKinsey reports for 2025. These are documents that cost $1,000 to produce. We get them for free, download the PDF - this is our raw data. Now let's go to Cloud. Why not GPT? Because Cloud is the best model for code and data analysis. Create a project, marketing analysis, isolate the context again so that nothing gets mixed up. Upload our pdf reports to the project. And now we ask, not just "make a summary," we ask: "Create an interactive dashboard" based on this report. Here I've set many parameters and described everything I want to get point by point. The prompt will be in the Telegram channel. Cloud launches the Artifacts function. It writes clean HTML, CSS code right in front of you. It turns dry data from PDF into a live interface. Click preview. Attention to the screen. This is not a picture, it's a working web application. The charts are interactive. Numbers pop up when you hover. We see investment volumes, trends, benchmarks. You can download this HTML file and send it to a client or open it for a presentation in a browser or just in cloud. It all works. And you can make edits by typing in the text. Right in the chat. We've turned 100 pages of boring text into a professional analytical tool in 2 minutes. Case 5. Content marketing. Imagine a news item about a contest in Telegram comes out. We need to quickly create a post with an infographic. First, facts. Let's go to Perplexity, type: "Give me 10 future features. New Telegram update." Based on the contest conditions, we get a structure, not fabrications. We have the meat for the post: Telegram Notes, role model. Copy this list. Now we need to package it visually. Let's go to cloud, drop it in the chat. And here's the main secret. We don't just ask: "Draw a picture." We write: "Visualize these findings as an SVG artifact. Make it beautiful for social media." This is an explanatory type of writing. I chose it. Why SVG? Because it's code. Cloud writes code that the browser turns into an image. Look, it has laid out neat tiles, selected icons, aligned fonts. The text is perfectly sharp. But that's not all. Click the download button. We download not a picture, but a vector file. Open this file in Figma or any other vector editor. And here's the power of this method. Every element is editable. Don't like an icon? Delete it, put your own from the pack. Want to change the brand color? No problem, go ahead. You've got a design base in 30 seconds, saving 2 hours of a designer's work from scratch. The scalability here is colossal. You can generate a set of 10-15 variants in a couple of minutes. This is objectively faster and more flexible than waiting for raster images to render. And most importantly, it's full control. You're not dependent on the neural network's randomness, but control every node in the vector. For commercial SMM design, this workflow is an absolute must-have. Case 6. Let's move on to system automation. We don't just write a prompt, we create our own jmbot. AI employees within GMI. I go to jmbots and create a new bot. Let's call it presentations business speeches. The most important thing is the instruction. We program its brain once. I write: "You are an expert in presentations. Your task is to highlight the essence, like in McKinsey, and remove fluff." Click generate instruction. You don't even need to create it in a separate chat. Everything has become much simpler. Here we set strict standard qualities and tone. Save. And the killer feature. You can share this bot with your team. You set up the logic once, send the link to your employees, and they use a ready-made tool without wasting time on prompt engineering in every chat. Test drive. Open our bot, upload a heavy global report to it. The bot already knows how to process it because it's embedded in its instructions. Create a presentation structure in Russian. It doesn't just copy text, it writes speaker notes, suggests which charts to use, and structures all the theses. The semantic part is ready. Now visualization. I ask it, create slides from the materials in this chat. Gini starts laying out layouts right here in the browser. It takes the structure it came up with itself and distributes it across the slides. And the main advantage of the Google ecosystem is the convert to presentation button. Click it. And we instantly go to Google Slides. Our presentation opens as a fully editable file. This is not a picture, it's text, headings, design. You've just saved 4 hours of manual layout and analysis. All that's left is to fix the details and you're ready to present. Case 7. Let's move on to the most complex task. Self-education. Suppose you want to learn machine learning from scratch. Usually, it's chaos, dozens of tabs, unclear articles, unverified information. We will build a systematic educational process in Notebook. I use the pro version to remove source limits. First, activate Deep Research. This is not Googling, it's launching an autonomous agent. Look what I'm doing. I'm not looking for articles in blogs. I give the command PDF files, textbooks, and technical documentation. Why is this important? PDFs usually contain academic peer-reviewed knowledge. This is pure fuel for our neural network. We immediately filter out marginal noise and SEO articles. Launch. And here you'll notice that the process isn't instantaneous. This is normal, unlike CHGPT, which gives an answer immediately. This agent actually goes to websites. It scans university repositories, checks file availability. Results are ready. Look at the quality of the selection. This is not Wikipedia, this is an MIT lecture, Python library documentation. Serious books. Data loaded. Now let's turn the library into a classroom. Reading everything indiscriminately is inefficient. I'll ask it to create a course structure. The system instantly gives me a course structure. We've turned chaos into structure. And now, the miracle. Multimodality. There's a lot of text, it's complex. Sometimes the brain needs a different perception. I go to the reviews section and choose not just audio, but a video review. We ask the neural network to become a lecturer. It analyzes the text, highlights the main diagrams, charts, and graphs from these PDF files. But just reading isn't enough, you need to test yourself. I ask it to create a test, quiz, or flashcards for memorization, and it generates questions strictly based on these textbooks. If I answer incorrectly, it will show me which chapter to look for the answer in. Look, this is a ready-made video lecture. >> Hello everyone. From recommendation feeds to facial recognition in photos. Machine learning. It's already literally everywhere. But how does it actually work? Let's take a look under the hood of this technology in this breakdown, and try to understand its key principles. And it all, absolutely everything, starts with such a seemingly simple, but actually very deep question. Can a computer be taught to learn at all? When we talk about a computer, look, there's a classic definition from Tom Mitchell. He's one of the pioneers in this field. It sounds like this. A model learns from data if its performance improves after considering that data. >> We have a host, we have a visual. Slides change in sync with the narration. The video shows the neural network architecture at the exact moment the speaker talks about it. This is the ideal format for complex technical onboarding. You can watch this during lunch, immersing yourself in the context without strain. I can do it by individual lessons, reinforcing the theoretical base. Plus, when creating it, I'll add more input data to the description, and it will make the video review even more accurate and detailed. It works quite well, not perfectly, but it's usable nonetheless. As a result, we had scattered files on the internet. In 30 minutes, we have a structured course, verified tests, and a video lecture. We've created a course product at the level of Coursera, but completely tailored to our tasks. This is exactly what the educational process should look like in 2025. Many students in foreign universities and even schoolchildren have already switched to this format. The NotebookLM service is gaining popularity, and I consider its integration into learning an absolutely correct step. CHGPT also has its own learning modes, but, of course, the leader in the educational sphere today remains Google with its solutions. Case 8. Now let's move from theory to real product production. Our task is to create an MVP of an educational course from scratch, based on real data, not on artificial intelligence hallucinations. We begin. Perplexity is our smart search engine. I don't ask it to just write text. I give it the role of a product analyst. I need a strategy for launching ML courses in 2025. Perplexity scans the internet in real-time, collects trends, and gives me a summary. Note that I'm not reading this here. I export the answer to PDF - this is the key point. We turn search results into a fixed document, a knowledge base for the next stage. Let's go to Notebook LM. This is our analytical center today. We upload this very PDF from Perplexity here. Now we have a closed context. The neural network will work only with this data verified by me, excluding fabrications. Now let's turn data into structure. I ask it to create a landing page architecture for our MVP. Notice, I'm not inventing blocks out of thin air. It analyzes audience pain points and uploaded videos and reports and offers solutions. We've got the perfect structure. Time to move on to writing code. I'll open GMI again. It's important to choose the right tool here. A regular chat won't quite work for us. I'm opening the canvas Holst specifically. This is a special interface for developers and copywriters. On the left, we have the chat with the neural network, and on the right, the code workspace. I paste our structure from Notebook LM and give the command "Layout a professional modern landing page for me." Based on this structure, use the Pro model for quality code. This is very important. We see the first screen, we see the course program, we see the tariff cards. This is no longer just text, it's a ready-made front-end site with a back-end. But this is the first version - it's just the beginning. Now I work as an art director. I don't like how the course structure design looks. I don't touch the code manually. I select an area or just write in the chat: "Make everything in a single modern style." Canvas understands the context, it rewrites only the necessary piece of code. If there's an error in the code, well, as in the example now, canvas often diagnoses and fixes them itself. Just click the fix error button. Also, if this pops up, you can refresh the page. It might disappear. Look, the design has been updated, neon backlighting has appeared, interactive elements. Everything works. We've got a fully responsive layout in minutes. Let's summarize. We've connected three powerful tools. Perplexity found the data, Notebook LN structured it and removed hallucinations. And GMI turned it into working code. Previously, this process took a week. The work of a marketing team, copywriter, and layout designer. Today, using this technology stack, one specialist creates an MVP product in 15 minutes. This is what specialist efficiency looks like in 2025. That one person can do essentially very, very much and very quickly. Let's summarize. I've shown this case using five specific services, but you can and should change these combinations for your tasks. Each tool has its own strength. Want to write complex code and create applications? Your best choice today is cloud. Need a universal multi-agent for any task? Use chat. Need to upload gigabytes of documents or an hour-long video for analysis? GMI is the undisputed leader here with its huge context window. Be sure to subscribe to the channel because very soon I will release a big review. Year-end review in the world of neural networks. All the giants have recently updated. This material will be relevant at least until the end of spring. You need to see this to stay on trend. So, like, subscribe, and also all the prompts and what I got, I've left in my Telegram channel. So go ahead, subscribe. It's a community without ads, I don't spam information, only useful content. And finally. Don't forget to implement what you've seen, but remember that for serious work, free versions are often not enough. If you need pro tariffs without complexity and with payment, the link is in the description. This is a trusted service with reasonable prices. No hassle with cards, just go, get access, and start working effectively. Write in the comments what you will try first from what was shown, or what combination you want me to cover in the next video. I read everything. Implement and do it systematically. See you in the next videos. Bye everyone.