Transcription
As any decent modern developer, I use neural networks. I have a certain set of tools and programs that I use to increase my productivity, write code faster, or think through some ideas more productively. And you often ask me, either on Telegram or here in the comments under the video, about which specific tools and programs I use and how. Therefore, let me tell you about it today, and there will be both native programs and web applications, and I will show you what they look like, and I will try to explain what I do with them. As you might guess, I use them not only for programming. And therefore, I have general-purpose tools that allow me to use myself anywhere and for any purpose. And this includes the same chatbots on the internet, the conditional Chat GPT, GRK, or Gemini from Google. I use them for creating texts, for example. And in this regard, I like Gemini more. It somehow knows how to invent better, let's say. I also use Chat GPT, for example, very often to think about code, because I like how it generates code more. But at the same time, if I want to think through my idea, look at its architecture, then GPT and Gemini work more or less the same here. I can use whichever I open first. But, as I already said, I can think through the architecture there and look at the meaning of this application. Does it have any meaning, or perhaps even study the market to find out if there is a competitor to what I am inventing. I have already mentioned that I write code there too, but lately this happens less and less. But thinking about the architecture and the components that will be in my application, which I want to write, that's yes, very convenient, cool, and any will do. I use Grok more for generating images, because I think it does images best. But even for chats, I don't always use an online editor. Sometimes I use local ones. either to see how well a particular model works, or to get a quick answer. For example, if I need some small snippet of code, then local models often generate it faster and better. And for this, I use a program called Ollama. It allows you to organize projects. In projects, you can create new chats, you can add various roles for the neural network and then quickly switch between them. It supports a huge number of different models. Here, right in the interface, you can conveniently download them. And Ollama also provides an API that can be used instead of OpenAI. But I use OpenAI specifically if I need a completely standard API, downloaded from the OpenAI website. And I install the model I need. And then I can direct my scripts to this model, so to speak, via a local IP address and see what happens. Thus, I also often see which models are better suited for certain purposes, and then I can use them for my tasks. Let's talk about web applications with local neural networks. And I use two of them. The first is background removal. It allows me to remove the background from myself if I need to make a preview for YouTube. And, as I already said, it is an absolutely local model. There are several versions of this model. And, accordingly, whether the file will be downloaded more or less depends on how powerful your computer is. I use the most powerful, this ultra-high model, and it cuts out the background brilliantly. And the second is Whisper, which allows you to convert audio to text in different languages. This can be done with local audio or video files, or you can dictate what you want to convert to text. I mainly use it to convert lectures for my students into text and then generate a text lecture from it. Well, and the last application, which is not always related to programming, is the application ID. I wrote it myself. It's even in the App Store for some money. And this application allows me to dictate the text that needs to be typed. It has a global hotkey. I place the cursor in the input field where I need to type something. For example, someone wrote to me in a messenger and I need to reply. I place the cursor in the input field. Then I press the hotkey Shift Command R. Recording begins. I say what I want to write. I press Shift Command R again. And after quick processing, the text is inserted. It's quite convenient. It saves a huge amount of time. Of course, you need to quickly reread what actually came out and maybe add some punctuation. Therefore, for a small text, it doesn't work at all and doesn't speed anything up. But if you need a lot of text, especially for prompts, yes, you know that a normal prompt to any modern artificial intelligence, to a large language model, to a chat, is a large, large amount of text that needs to be typed. So, for prompts, I use exactly this program of mine. And punctuation is often completely unimportant there. It works completely locally. It uses a model from Whisper and is written in Swift and C. It can also translate to English from any language, if needed, and this is a limitation of Whisper, and it can save everything you dictate into files if you want to return to it later. Well, now for programs that I use only for development. And I'll start, perhaps, with the console. This is Warp. It's a perfectly normal terminal. It has a free version, which I use. There is a paid one. And this terminal allows you not only to perform functions that you usually perform in a standard terminal or in some iTerm, but it has a built-in neural network, remote, of course, and therefore, when you press the standard hotkey Command Y, as in all other programs, this neural network chat turns on, and you can type what you need. I most often use it to make some complex grep or JQ. That is, I need to take a huge amount of text data that I have and find something there, or select something, or maybe it's JSON, and I need to work with it. It's very difficult and time-consuming to come up with all these complex constructions yourself, so the artificial intelligence built into the terminal helps a lot. Well, and code editors, where would we be without them. And I use four code editors with neural networks. The first is, in fact, Xcode. I have a plugin from Cursor installed there, its free version. And it helps me search the codebase, as it cannot make changes itself, create files directly. It can only suggest what needs to be done. Or at least, that's what it seems to me it can do. I use it to look at the codebase, find what I need there, and maybe suggest options and solutions that need to be done to do something. I also have AntiGravity, and I really like this program, also the free version. And I have already written two hobby projects using AntiGravity from scratch completely. And I really like how it generates code, how quickly it does it. It has a free version. In the free version, there is a decent selection of different models that can be used with different limitations. And I use, of course, Sonnet whenever possible. Well, and if the limit runs out, then Gemini 3 Fast Flash. In fact, Windsurf itself is a separate application, also a clone of Xcode. I don't use it very often, mainly for some small edits to existing applications. The limits there are slightly better than in the free Cursor or AntiGravity. Well, as I think, and it works well. They have their own model. However, I tried to create programs from scratch, I didn't really like the result. Well, but if you need to quickly fix something, find it, do it, and do it all automatically, then it's perfect. And finally, Cursor is my main code editor. I have a paid subscription, and I use it both to write programs from scratch and to make complex changes with a huge number of files and modifications. I don't do everything automatically in it, especially if the task is complex, but it does most of the work and writes 90% of the code for me lately and then corrects my comments for me as well. I haven't tried using Codex, which works in the terminal yet. And I'm curious what will happen if Codex is combined with Warp. A neural network with a neural network, such a super-complex agent or something else. It will be fun to see. Well, and so, these are approximately all the applications, web applications, and tools with neural networks that I use. Of course, sometimes I use chat, online chat from Quentin or from Deepseek, but lately this happens very rarely or not at all, so I didn't talk about them. Well, in general, if you think that I am missing some tools here to work better, then write in the comments. Maybe you think there is a replacement for what I use, also write in the comments. Like this video if you enjoyed it. Subscribe to the channel so you don't miss my other videos. Well, I'm not saying goodbye to you, this is the internet. See you soon.