Transcription
Most people use neural networks too primitively. They simply throw a bunch of text into one chat, ask for a summary, and get absolutely bland, boring water. If this still works for simple tasks, it's absolutely not for truly deep analysis. You will never find cool ideas or hidden trends if you dump all the information into one pile. The neural network will simply drown in this volume, lose context, and give you banality. Everyone wants to press one magic button. But the truth is, artificial intelligence needs the right conveyor belt. In this video, I will show you my deep research system. We will take three absolutely different sources: dry theory from the internet, live correspondence of people in Telegram chats, and real experience from forums. I will show you step-by-step how to extract the essence from each of them separately, and then combine them into one powerful final report. You will learn to find information that your competitors simply do not see. On communication. Let's figure it out. Let's start with my mind map, which I have prepared for you. It may look a bit chaotic somewhere, of course. This is my stream of thoughts that I was laying out, I deleted some things, added some, but in the end, what turned out, turned out, and I like it. And we will start from the end, so that you understand how this system works at all. When we, well, not us, but most users try to find something through a neural network, they do, well, often deep research. And this function is available in almost all popular neural networks now. GPT, Dips Gemini, GRK, and so on. And after the neural network gives them some answer, they immediately take it and start processing it further somewhere, uploading it, and immediately accept it as truth. But what's the catch is that each neural network is trained on its own data. Each neural network has access to different search engines. Well, because there are certain agreements between companies. That is, for example, one company partners with Google, another with Bing, a third with some other search engine. And it turns out that even if we send the same request to different neural networks, they will give us different answers. Well, it's clear that somewhere there will be something similar, somewhere it will differ, somewhere there will be contradictions. And our task is precisely to create a general synthesis, yes, several different studies, yes, which I talked about at the beginning, in order to then get such a grand final report, where the information will already be synthesized, yes, among themselves and there will be no contradictions or hallucinations. That is, we will minimize them. And now you will understand, look, I will break it down in order. I took three independent sources. The first source is pure neural network research in a general order, yes? That is, we compose a request. I will now show how to do this through third-party services, which are also cool. And after that, we distribute this request to three to five neural networks. Then we get reports from these neural networks, and we have to synthesize them. This is my second step, synthesizing these reports. That is, we find some overlaps, similar points, contradictions, differences. And we get such a final report on the first run. The second run is my export of Telegram chats. That is, we find thematic Telegram chats. By the way, I even have my own application, it's a Telegram exporter. We will also consider it in the process of this video. I will show how to use it and where to download it. That is, we export channels, chats, anything specifically on our topic. And then we also do a synthesis based on these data, it turns out. Well, and we also get a report. That is, we are just talking in broad strokes now, then we will delve into each step separately. And third, we take and do the same research again through several neural networks, 3-5 pieces, but already specifically focusing on various forums, discussions, and also synthesize them, get a report. And then at the output, we have three reports from three independent sources, and we synthesize them together again. And we get such a clean OS, I called it an operating system, with which we can then work, further load into some separate neural networks or into LM notebooks. to load 1, 2, 3 reports and work with this information. That is, we don't just throw something in, yes, well, what the first neural network gave us, but we carry out a certain chain of work, after which we get a clean study and can then do, well, whatever we want with it absolutely. Well, there are also ideas, I will voice them in the process. Let's start in order. The video probably won't be super short, because there are also related topics that I would like to touch upon. I want to make it a truly comprehensive research. The first thing we need to do, as I said, is to turn to the first chain, that is, to make a general request to neural networks on our research topic. But an important point, yes, we cannot make a request out of thin air. We need some input data, context. For this, I have compiled this small brief. You can find all the links in my Telegram channel, including the link to this mind map. The brief, in which we fill in some columns, is the topic, key queries, types of sources, depth, time horizon, and expected output. This is a draft version. Don't worry, you can write in your own words here, you don't need to adhere to any prompting techniques. We will then process all this information from the brief through a separate service. Look, I took neural networks as an example. That is, I have already gone through this whole path. I took the topic: what problems do beginners face when working with neural networks in 2026? Key questions: where to start? How to pay for foreign services? How to register for Google services, where to find news about artificial intelligence, how to start coding, which neural networks to pay for, and so on. Types of sources. Here I covered everything. That is, this first study covers everything. That is, articles, forums, correspondence, open sources. YouTube research, social networks. In general, the more you throw in, the better. Depth, deep research on the main pain points of beginners in the world of artificial intelligence. What are people looking for now, googling, watching, what is gaining the most views in this area? YouTube, articles, and forums. Time horizon, the last 12 months, but prioritize the freshest and most relevant information, starting with the last month, then 2 months, 3 months, and so on. And expected output. I decided to make it a detailed report plus sources plus a table with topic ranking. The table should indicate the topic's hotness, justification, where the information was taken from, and links to sources. This is a small brief, filled out in about 10 minutes, I would say. We just dump all our thoughts here. With this methodology, you can research absolutely any topic for any of your tasks. For example, learning. You want to learn something, and you need to gather information, where to start, what tools to use, and so on. This is also a good option. Second, conduct a competitive analysis, that is, see what competitors have, what is on the market, where it hurts now, where it doesn't. Well, you understand, this system allows you to research absolutely any question if you just articulate it correctly from the start. Next, we move to a service called Prompt Cowboy. Again, I will provide all the links in my Telegram, if you don't want to search for it separately. It looks like this. Basically, you will need to log in here through your Google account, and you will see this window. What does this service allow you to do? It allows you to convert your stream of thought into well-structured prompts in different categories. Here we see standard research, writing, planning, agents. Well, I won't list them all now. And precisely, we need to get the research prompt from this service. The service is paid, but partially. There are 10 requests per month using the modern model, that is, Opus 4.6, which is an advanced model on a neural network. As soon as you make 10 requests, that is, you send them to the chat here, it switches to the Haiku model, which is a lighter one, and it will also work without limits, just with a different, well, a weaker model. If you want to use the latest modern models without limits, the subscription here, I think, costs, well, it showed me eight dollars somewhere, fifteen somewhere. I use the service constantly, I run all my requests through it. Initially, let's go back to this. By the way, I already did research here, we will get to it. We will take this small template and write: "I need to do deep research, here is the context." We select research here. And then we just very carefully copy the information from our brief here, one by one. Let's do it now. I'll just start, then I'll pause and continue so you don't have to watch for too long. So we take the topic like this and copy it from the brief. Next, we take the key queries, also a colon, and copy from the brief. And we do this with all the columns. Let's insert a pause and then I'll get back to you once everything is filled in. I have filled out the brief, just transferred all the data from the table. Well, not the brief, but this request to our service. And now, that's it, I've left it as is. I need to do deep research. And I click send. Now you will see the magic before your eyes. It will create a research prompt for us. So, we will now send this prompt to different neural networks. Here it wrote: "You are a senior research analyst specializing in digital trends and user behavior in the field of artificial intelligence." Blah blah blah, it describes everything here. And on the left, I advise you, when we have filled out the brief and uploaded it here, precisely for our first request, answer three simple questions on the left. So, who is this report intended for? Well, let's say for myself. What is the technical audience level? Well, the technical audience level. But here we can write something from ourselves, that I am doing this report for myself, in order to later find interesting topics for content on YouTube, for example. Send. And the next, the third. What is the format of the final report needed? A full report with methodology and all sources. 15+ pages. Then click the Improve Prompt button, and it rewrites it with our additions. And we will now see exactly how the basis for a YouTube channel content plan. And that's it. What do we do with this request next? Let's open our mind map while it's writing and look. The step is, well, we have to send this report to different models now. That is, choose from three to five. I did the report in all popular models. Well, I think it's a larger sample, the better. This is Perplexity. This is through Google AI Studio. Well, there I chose GMIN 1 Pro. Gemini. You can do it through the usual, yes, service. Like we go to Gemini, type it in search, through the website. Then it's Deepsek, Grog, and Chat GPT. So I have five neural networks. And I upload this report into each of these neural networks. Now I will show everything. Well, let's go through it so it's clear. Okay, it wrote it for us, we copy it. And now let's go to Perplexity in this same tab. Then Deepsek, then GRK, after that Gemini. And, what else? Chat GPT. We go into each of these neural networks. Somewhere here they ask us to confirm something, to log in, to authorize. Okay, here we select deep research. Let's see if it's available to us. Yes, it's available. We paste our request and send it. Then we open the web version of Deepsek. The same here. We log in, select Google account. Our account. Continue. While it's loading, let's also authorize in Grog. So, to be faster. Also Google account. Continue. Okay, so, Deepsek login is complete. We select deep thinking and smart search. We paste the same request. Send. Let's authorize in Grog. Let's wait a bit, while we do it in Gemini. Okay, I've already logged into Gemini. By the way, for those who are interested, I use this browser VPN, and it works perfectly with all services and neural networks. Here's my paid version. Let's send to Grog. We select expert, deep thinking here. The same Gemini in the Tools section, select deep research, send. Here we can select all resources. And here the model is not Fast, but Pro. So, Pro. Well, it seems Pro is by subscription. Well, let's take Thinking, yes, the model and select Deep Research. Send. And Chat GPT, we also select deep research. Send the request. Usually, Chat GPT also asks some questions, but let's answer them. I'll do that now and wait for all the neural networks to finish their reports. All neural networks have successfully completed deep research. Let's see how it looks. Here's Perplexity, yes, it made a separate window with a report. I clicked the export button and downloaded it as Markdown. I got a file. I right-clicked on it in the folder. Open with text editor. And from here, well, I copied the text and created a folder called Deep Research on Google Drive. And additionally, in this folder, I created another one called One Step, that is, the first step. And here I created these files. Initially, since I took Perplexity, I created OneStep_Perplexity and pasted here, in this file, what I copied from this text file. The second step is Deepsek. Here I just copied, also created a OneStep_Deepsek file and put the entire report in. Next is Grog. I copied and also pasted the entire report here. After that, Gemini. Here we also need to click Share and Export, yes? So, let me close it. The report looks like this. We click open and here copy contents. And I also paste it here as a final report. And the last one. So, this is, well, not the last one yet. Here's Chat GPT. It's a bit more complicated here. That is, when it makes the report, we need to click on it, it will expand. We scroll all the way down, click this button, download it in docs format, and then upload it here. See, I uploaded it in the bottom right corner, into this OneStep folder, and renamed it Chat GPT. So that everything is in a unified format. After that, we get five independent studies from different, or rather, different neural networks on the same request. We will work with this further. How, well, what is our next step? We need to download all these files in MD format now. We click the Download Markdown button here and do this with all files. Download. Then Grog, the third file. Then Gemini and Chat GPT. Okay, we have downloaded five MD files, and now we need to synthesize them. For this, I have a separate prompt, I prepared it. Of course, you can improve it yourself, supplement it. Maybe someone has their own methodologies. It looks ordinary, well, that is, it is designed to find some contradictions based on all five studies. Here it says, absolute consensus, that is, all solid facts or statements that all uploaded reports agree on without exception, analytical base, points of divergence, marketing noise and water, blind spots, and so on. Therefore, how will we do this synthesis? Since the report is not small, I clearly want to ensure that the neural network can handle it and work with it. Here we have a parameter or definition like the model's context window. I won't go into too much detail on this topic now. In my Telegram channel, well, here I made a link to a YouTube video and to Telegram, because I have one on YouTube, another in Telegram. Here I attached my video, which is an excellent super base on neural networks. Well, it's really deep, I recorded it myself. And also a detailed lesson on model context windows, it's also about 20 minutes long, very cool, detailed, where I thoroughly explained everything, what a context window is, how it works, why it works that way, and for what. But briefly, I will note that we will use the Gemini 3.1 Pro version for synthesis, because, firstly, it is a smart model, and secondly, it has a context window of 1,857,600 tokens. What does this mean? This means that, well, you can upload quite a lot, maybe about 30-50 of these reports, yes, maybe a little more. We will use this model through Google AI Studio. And we, I will now show how to access this service. And as a bonus, I wanted to show how to connect a foreign card to use the API and get a test $300. Okay. Let's switch then, let's type Google AI Studio in the search. Different results open. We open the first site. Probably, it will ask you to log in through Google. You most likely log in, and you will see this window. You need to click on the Playground section and in the upper right corner, select the Gemini 3.1 Pro preview model. Here we click the get button so that, well, if you want to work a lot through this service, you will have to link your foreign card to be able to select your key here, select key. And after that, your key will be used and your test $300 will be debited. How to get it? Click get API in the bottom left corner. Here, if you don't have any keys created, then click create API, give it a human-readable name and click create. It will also appear here as a line. See, I have already created my key, named it RX AI app. And all I need to do here is to link billing. By the way, I probably chose not a very good account. Let's open an account where I haven't linked a key yet, to show the button called Setup Billing. Here it will be, if your card is not linked. You click on it, and now you will need to link a foreign card. Where to get this foreign card? Let's open Telegram and I will show you a solution that I found, tested, and it works not only for me but also for many people. It's called an application, let's call it that, or a bot. Zarubro Bot. You can either search for it yourself in Telegram, or get the link in my Telegram channel. When you click on it, this window will open. On the left, click on the Cards section, and you will see your card balance, total balance. I will now show you step-by-step what needs to be done here. First, you need to top up the general balance. It's called general. You won't have any cards created here. You click top up balance and choose a convenient method for you. It can be SBP, or those who use cryptocurrency USD stable coin. Rubles can also be selected here. This is probably some alternative method. Then Altcoin and coin. Please choose any of the methods, top up, and as soon as the money reaches your balance, if I'm not mistaken, the minimum top-up here is $10. If you are converting to rubles. Well, let's take with a margin, 1000, 1200, even 1300 rubles. Aim for this amount, because there is also a small commission when topping up the card. After we have the account balance, we click the issue card button. We see that it says $8 here. When we have topped up the balance, we click the buy button. Specifically, this Virtual Card. And it appears at the bottom. Its balance will be zero, because in this bot, the general balance and the balance on the card are separated. When we have the card, we click the first green button and top it up from the general balance. We enter the amount here, it will be in dollars. Well, let's write, for example, 10. And we see that there will also be a 1.5% commission from the balance you want to top up. Click the confirm button and wait for some time. If, for some reason, some information is not loading, click the refresh button in the upper right corner, and the application will simply refresh. And there, well, it takes about a minute for the balance to appear on the card. Next, to use the card details, you need to click this pink button. I won't do it now. I will show a screenshot separately so as not to reveal my card number, CVC code, and expiration date. And the billing address for filling is also written there, that is, the country, city, zip code, and so on. And the third button is your transactions, how much you have spent from this card per month. Well, we see, in this case, I spent $210. That is, I use it actively and, well, I recommend this service for a reason. Let's switch back to Google AI Studio, and with our screenshot, we will fill in all this information together. I took a screenshot, I didn't. You see, there are dark fields at the top. Here I have three fields written. These are, as I said, the number, expiration date, and CVC code. I will show you later where to enter them if needed. So, let's look here. If you don't know, for example, what country this is, just send this screenshot to, for example, Chat GPT. Let me show you clearly. Let's take this same chat where we did the research. And here we send it and say: "What country does this address belong to?" Well, just in case someone doesn't know. And we see, yes, it's the USA. Yes, you see, if you ask which country the specified address belongs to, it's the USA. Everything is fine. Let's go back to Google AI Studio. So. Here, let's click Setup Billing again and then select United States. Then click Agree. And now we will fill in the data. I just want to show step-by-step so that no one has any difficulties later. We also see that when adding a new card, you get $300 for free to use AI, well, for your account. In fact, you get $300 to use the Gemini 3.1 Pro model and for generating photos in Nanoban Pro. I will show this too. Click the first button, Add an address, and here select not organization, but individual, enter the name indicated on your card. In this case, if we look at the screenshot, it's my name here, USY. We enter it here. Then, to not bother with the rest of the lines, let's do it very simply. I take a screenshot, I take a screenshot. Then I take, go to Chat GPT and send him these three screenshots. The first is with my card details. The second is the first field. And I can't attach the third. Well, let's see what we have here. State and zip code. Okay, send such a simple request to Chat GPT. I struggled a bit with Chat GPT. In the end, I wrote to him: "Just take the data, fill in the fields according to the screenshots plus state and zip code." He wrote everything for me. Then we take this and just transfer it, right? I have already filled in the first line. Then street address. Let me copy it from here. Then we skip this field, it's not mandatory. We indicate the city. Then our state is Alabama and the zip code is 36116. And click save. No, thank you. And you see, the contact information has been pulled up. And then we simply add a payment method. We select add credit or debit card and enter very simply, well, the card number, expiration date, and code. Everything, click save. And then after that, there will be a button, like, well, finish setup. Well, and then proceed with payment. Okay, the card is linked, and then our key, well, the key becomes active, and we can use it.
to use in work. So, let's take this project now, let it be, uh, free even, because there is an opportunity to use these models for free, but if you connect a key, as you can see, here you get access to both Nanobanna Pro, and you can generate a huge number of images there for these 300 dollars and text, including. I've been sitting here literally for days on end and have only spent 100 dollars out of 300. That's all. Next, we take all our reports, which we exported in MD format. This is the first one – Perplexity, then Deepseek, then Grok, Gemini, and Chat GPT. We see that our context window, as I said, remember, is a million. Here. And we currently have 72,260.5 occupied by these reports. Now let me switch back to my mind map and after that, I'll copy my prompt that I showed. Okay, now let's see that GMIN 3.1 Pro is selected here. We can also, by the way, set the URL context here as well. Just in case, in case it needs some context for links, and press the Run button. Okay, we have five reports. Uh, a prompt for synthesis. And now, at the output, we should get, uh, a detailed final report, uh, with overlaps, some disagreements, and conclusions. Let's wait. Okay, it's started writing the report, synthesis, cross-analysis of barriers to entry for beginners into the AI sphere 2026, absolute consensus, analytical base. Well, that is, everything follows the output format clearly. Next, the points of divergence, it will write now. Uh-huh. Yes. And let's wait until it finishes writing everything. Then, when it has finished writing the report, we scroll to the very top. Here we click on the three dots. Uh, Copy As. And it says there, as it was written, yes, Markdown. We need the Markdown format. It's read best. Well, it's more convenient to work with. And neural networks read it better. And we go, uh, to the main folder. Click Create Google Document. And here we write One Step. One Step. Well, in English, report. Well, I forgot how, in short. Let's just write report for now. And paste all this information here. That is, this is our first report, which we received based on five studies, that is, five different independent studies of neural networks. Next, in the step, we have the following – to find thematic Telegram chats using my application. Here. So, we've finished this step. Next is the next chain, Telegram Export. What's the idea? Most likely, on the topic you want to research, there are already some chats, thematic channels where people discuss, publish, post, and share their expertise. Our task is to collect all the information from there and then also find a common synthesis between all these chats. I've highlighted different services on my mind map, well, some kind of search engines for Telegram. You can go to them, they look very simple. That is, Telegram, something like that, right, type neural networks here, and it will give you specifically posts from Telegram, from different channels and something else. Here we see a public, Telegram, Voiceat. Well, and I've selected a large number of such services. And our task is, using these services with such keywords, or, for example, the same popular TGStat, I've added it here too, to select channels, but preferably at least one or two chats where people are communicating. After that, when we've found them, and we need to subscribe to them from our personal account, we go and download my application, which I developed, called Telegram, well, Export. What does it allow you to do? You log in with your account, and then you get some export settings. Let me open it first, and then I'll tell you during the process where you can download it. It looks like this. We log in with our account. And when we subscribe to any new channels, we must click the "Update" button. They are pulled up. We type them in the search here, search for each one separately. And, for example, let me take my folder now, select it. You can also export entire folders here, by the way. That is, we can put all these chats and channels into one folder and make an export of the entire folder. And then, uh, let me select my channel. I have the option to choose a period. And if you are exporting a chat, I advise you to do it not for all time, but, for example, for the last 3 months. Otherwise, if the chat is huge, you will have a large amount of text, and it may not even fit into the context window of Google AI Studio. We must take this into account. Uh, then you don't need to set anything here specifically for exporting chats. Here you can separately watch my video in my Telegram. Let me show you. I recently posted an article in my Telegram channel, on YouTube. You can watch it, where I talked about this application in detail, or in my Telegram from February 7th. Here I posted a download link, an installation link. So, there is all the detailed instructions here, so, um, well, watch, study. Now let's return to the application. I press the "Export selected chat" button at the bottom. Here. And I chose the last 3 months. Uh, then I press "Yes". After that, I choose the folder where I need to save all this. Let's say to the desktop. And then the export process starts. Well, if the chat or channel is small, it will be very fast. If it's larger, it might take from 5-10 minutes, but that's in rare cases. Okay, we see here I have two files. This is, uh, pure JSON format, with all sorts of brackets, commas, and explanations. Well, just in case, in case someone needs it and wants to work with such a format. And we get this clean, cleaned-up Markdown format without any extra elements. All the links used by the author are included here, for example, if he posts something in the channel, and all his posts are step-by-step from the very beginning for the last, uh, well, in this case, for the last 3 months, if I had chosen all time, then it would have been from the very beginning. And we collect five of these. Next, we go and open. So, let me move to my previous account. Uh, open Google AI Studio. When we have collected five such thematic chats, we upload, uh, these MD files. Let me upload one file now. Chat one, uh, I made chat 2. Here, let's say chat one, for example, let's imagine this is the second chat, then chat two, then chat three or channel three, it doesn't matter, and, for example, the fifth. Well, you can have more, you can have less. Well, let's say the sixth. And now we need to paste the prompt again, which will allow us to synthesize all these Telegram chats. Let's copy it now too. I've pasted the prompt, I'll attach it too. Well, it will be in the mind map and in the channel. You are an analyst and social listening specialist. I have uploaded this raw export of Telegram chats or channels, chaotic live conversations of real people. Well, the task is described here, as well as the structure, uh, there, top five burning pain points. Well, here you can generally say, uh, there, top burning pain points, not necessarily top five. Folk solutions, life hacks, main questions, emotional tools, trends. All of this, notice, my new chat is new. We see that 514,000 tokens are already used here. Yes, at the same time, I attached, like, five chats, so, you see, somewhere I have 40,000, somewhere 140, if it's, well, a direct chat. Okay, press the send button. Let's also give it URL context just in case, so that it can have internet access and read links. And we expect the final answer and our report from it. It has written a full report. We also copy it. Click on the three dots, Copy As Markdown. I created a new folder, T Step, because this is already the second step of the research. Here you can upload the chats that you used for the research. Well, just in case, yes, in case you want to refine or, uh, well, improve something with this information later, so as not to lose it, in general. Open, uh, our document and paste here, uh, our second report. Okay, we've done two key things at this step. By the way, I have a download link here on my mind map, uh, for this Telegram Exporter application. We've completed the second step. Here, upon request. Here, if you hover over the icon, the prompt is shown. Well, I'll also post them in Telegram separately. I thought it was inconvenient to copy here, because they are split in half and then you have to select and transfer all of it. In short, it's easier to make it a separate file. Uh, everything is fine. And the third step now is our task to research through neural networks, uh, well, to do research on forums. That is, specifically, the first one was a general request. Now we are specifically focusing on forums. I also have a prompt like this here. Uh, you can also refine and adapt it for yourself if needed. You are an expert in finding information on forums and communities. Your task is to find real discussions, opinions, and experiences of people on a given topic. Mandatory sources for searching. You can add them, change them, delete unnecessary ones if you have some specifics in your niche. What you know is that your topic is discussed specifically on these forums. Then the response format and search topic. You can write that the topic is attached, let's write it in the report. And so as not to write by hand, let's also take our five neural networks. Well, you see, if you have a free plan on Perplexity, then you won't be able to do deep research in this case. Well, let's do a regular one, and attach it to him. Let this be our, uh, final report. Let's download it in Markdown mode, or rather, in format, and attach it. And now our task is also to duplicate all this for five neural networks and wait for them to give us an answer. I won't show the whole process again. Now I will create all these five reports, upload them to the folder, uh, final versions, and then we will return when, uh, we will upload them and synthesize them. I've already tinkered with some here, well, like, to skip steps and not repeat myself. I've also made, uh, three reports, or rather five reports, yes, at the third step. So, why is there duplication here? Uh, duplication. Well, in general, there are five reports here. These are Deepseek, Chat GPT, Gemini, Grok, and Perplexity, as in the first step. Then I sent everything to Google AI Studio, attached five reports, and wrote a specific prompt. It will also be available to you, please. You can slightly refine, adapt, or modify it through the same prompt cowboy that I showed you today. It synthesized information for me, made practices, barriers, and technical shifts in the use of LLMs. Well, that is, it synthesized information from forums. And I copied all of this and created another document. Uh, the third step-report. Okay, now I have three reports in hand, if I return to the main folder. And now my task is to synthesize these three reports and get one large detailed final report. And now I will have four, you could say, main documents with which I will work further. And, uh, well, for example, through the same neural network, where I will send them, or through NotebookLM. We will also analyze everything now. And another important point, look, I forgot to mention it too. Let's go back to the step when we were collecting information from forums, when you send a request to a neural network, be sure, be sure to specify the topic itself with the research request, otherwise it will search for related topics within the file, like, uh, there, there, limitations of access to neural networks, some problems with prompt coding. That is, well, it will extract subtopics from them. We have one general topic that we are researching, so, at step three, when we collect forums, be sure to specify your initial research topic. Well, and as context, it's up to you, you can attach, you can not attach some document. Well, I attached it, but in general, you can also not do it, yes? Or or as you wish. Uh, we get the grand synthesis. Evolution, barriers, and user split in the Russian Federation, meta-analysis. Well, and so, it will format all of this now. And then we can, since in this chat there is only the context of these, well, three plus the current final, that is, four reports, we can continue to communicate directly in Google AI Studio and, uh, then based on our research, ask questions and do something, for example, a content plan, I don't know, topics for videos, or some automation, if you researched the topic of how to create some automation there. Well, in such and such a direction, in such and such a business, and with such and such tools. Here. So, then there is communication with the neural network, it's already based on, uh, such synthesized materials. Let me go back to the mind map. We'll see that we've gone through everything. That is, we researched forums, got the third report, did cross-synthesis, and here, well, everything was put into a general one, we got the final one. Master prompt, by the way, for NotebookLM, I have also prepared a master prompt here separately, which you can insert there. Well, let me show you, so as not to be unfounded. Let's go to the NotebookLM website. So, you create a new notebook, then click on the three dots. No, not on the three dots. Settings, somehow it was here. In general, maybe you need to attach a source, right? Well, let's attach, for example, our third report. And, uh, here it appears, you see, the "Configure Notebook" button. And here we can, uh, set the communication style, your own option, and paste it here. Well, again, you can adapt it for yourself. Master prompt. And I have also prepared useful extensions for NotebookLM. Uh, there are very interesting ones here, by the way, take a look at them. Such a table. That is, this expands the possibilities of working with NotebookLM. That is, this is an archive of chats. I really like the first one, Cortex. There are a lot of possibilities added inside this tool. Different sorts, folders, and something like that. In general, there's everything there, there's even a detailed video. Then folders for sources, saved prompts, deduplication. Well, here are some words, if anything, complex. There, inside, when you follow the link, there is a detailed description of what it does. And there are many such interesting extensions for NotebookLM. I've looked at all of them, checked them all, they all work. Also, there is another interesting extension – Perplexity to NotebookLM. That is, it allows you to export, well, threads, that is, the same chats, your discussions directly into Perplexity, saving the sources. So, that's also interesting. And this is the video that turned out. So, uh, the idea is that when we want to research a topic, we don't just throw in a request and do deep research in one neural network, but we do a kind of, uh, scattering seeds, let's call it that. That is, first we take one independent source, do research in different neural networks, then we take Telegram, some other independent source. Well, by the way, you can replace Telegram with anything, just some other independent source on the same topic. Then you take some forums, and this is like a third independent source, again in different neural networks. Then you synthesize each of these reports separately. You get three reports. And then you make a final grand report from these three, I called it that. Okay, you have a complete research in hand. Then you can study this report, read it, and draw some conclusions based on it. Therefore, if you want to implement this into your work on a permanent basis, you will need to go through all this step by step, uh, well, like, with me on the video, what I showed. Well, and if the video was useful, be sure to subscribe, like, and leave your comment. Maybe my system should be supplemented somehow, right? Or are there any other more interesting approaches and methods that allow for even better results, because I value quality, right? I am interested in such research solutions, works, so that at the output from the neural network, you get not just hallucinations and water, but truly, uh, serious and useful material, with which I can further, uh, well, interact, make decisions based on it, and build plans, right. Well, again, including my own head and criticism. As is tradition, thank you all for watching and see you in the channel.