Skip to content
Back to all articles
By TJ SoftUpdated:

AI Implementation – Table of Contents of Articles

Wdrażanie AI spis artykułów

Below you will find a collection of summaries of my articles on implementing and making practical use of artificial intelligence. Each text was written with a specific aspect of working with AI in mind – from the basics for complete beginners, through tools that streamline everyday work, to advanced content-creation techniques and comparisons of different AI ecosystems.

I have arranged the articles chronologically, which makes it possible to see how dynamically artificial intelligence has developed over the past two years. The oldest texts, although still valuable in their fundamental principles, already require updating—particularly those concerning specific features that are evolving rapidly. The latest materials reflect the current state of the technology at the beginning of 2025, when tools such as ChatGPT Deep Search, Gemini, and NotebookLM had reached maturity, becoming true “game changers” in everyday work.

Table of contents

General Guides

How to Start Using Artificial Intelligence in 2025 If You Haven’t Used It Before?

03/12/2025

The article is a comprehensive guide for people who are just beginning their journey with artificial intelligence in 2025. It presents three main reasons to learn about AI: speeding up work and saving money, understanding your position in the job market, and the possibility of finding a new career in the AI industry. The most important thing is to understand that using AI is now as simple as searching on Google—it does not require creating an account (ChatGPT and Grok work directly in a browser), installing special software, or knowing English. You can start with practical tasks: explaining unfamiliar concepts, correcting texts, searching for information without knowing specific keywords, uploading photos for analysis (e.g., whether you can repair a door handle yourself), or generating graphics and videos. Particularly useful are features such as Deep Search for comprehensive topic research, NotebookLM for working with multiple sources simultaneously, and ChatGPT’s study mode.

Three principles are key to effective prompting: write naturally as you would to a person (you don’t need special commands), provide context (the model doesn’t know your situation), and don’t try to create the perfect prompt right away—it’s better to iterate and improve it in subsequent messages. Safety is equally important: models can “hallucinate” (make up facts), so always verify information, don’t ask leading questions that suggest a specific answer, and don’t share private data. In practice, AI works like an all-knowing assistant from the other side of the world—it has vast knowledge, but it knows only as much about your needs as you tell it. Useful tools include ChatGPT, Grok (generates videos with sound for free), Gemini (with the Nano Banana Pro model for charts), Claude, and the Polish Bielik. After a few simple attempts, AI becomes a natural tool that saves time in everyday life.

Presentations in Google Gemini

18/11/2025

In my article, I explain why creating presentations with Gemini is a real game changer for me (I’ll be overusing this term). This tool eliminates the tedious, mechanical work of arranging slides, allowing me to focus on the substance.

My key takeaways:

  • Canvas mode: I create the presentation directly in Gemini by asking for a slide format. A huge advantage is that I can open the finished project in Google Slides and edit it freely.
  • Presentations “for myself”: Thanks to AI’s speed (a dozen or so seconds), I started creating slides as personal notes or study flashcards. This is a completely new way of using this format, which was previously too time-consuming.
  • Disadvantages and limitations: I notice a certain repetitiveness in the layouts, as well as text overlap errors, which I can nevertheless fix manually in a few seconds. I am most vigilant when it comes to images—Gemini provides sources, but they are not always royalty-free licenses. So if the presentation is to be shown publicly, it is a good idea to replace those images.

Artificial intelligence does not completely replace humans here, but it provides a solid foundation, saving time on the most routine stages of the work.

Comparison of Deep Search Tools

07/03/2025

In my test, I compared the latest Deep Research features, which revolutionized the way information is gathered in 2025. I checked how ChatGPT Deep Search, Gemini, Grok, and Perplexity handle a comprehensive analysis of the Polish AI market. Each of these tools represents a different working philosophy:

  • ChatGPT Deep Search: A real “game changer” for me. It generates the most comprehensive and best-structured texts (up to 13,000 words), saving 2 to 4 hours of work on a single article. Currently, the other tools described below also handle longer texts well, but ChatGPT is still the best for long-form content.
  • Perplexity Deep Search: Stands out for its precision and specific business data. It performs best when searching for informational graphics and analyzing institutions, although it sometimes has trouble distinguishing between companies with similar names.
  • Google Gemini: The fastest in the lineup. It is excellent for quick summaries and searching YouTube, but its analyses can be superficial and often rely on poorly readable lists.
  • Grok (xAI): Thanks to its integration with the X platform, it is unrivaled for tracking trends in real time. It offers considerable freedom (no censorship), but requires vigilance due to occasional hallucinations in links.

Although these tools can independently perform 70–80% of the research work, the key conclusion from my tests is the need to verify sources. Deep Search does not replace critical thinking, but it drastically shortens the path from asking a question to receiving a ready-made, multi-page analysis.

How to Create Texts with AI

03/03/2025

In my article “How to Create Texts with the Help of Artificial Intelligence”, I show how to use AI (ChatGPT, Claude, Gemini) as a powerful assistant that does not replace thinking but dramatically shortens the creative process—from 12 hours to just 4. This 4–12 principle is very important: a text written in a few seconds, from scratch, with a single command, is not particularly useful, and publishing it is merely a waste of both your own time and your readers’. If we spend several hours creating a text and analyze the article and check its sources, AI simply becomes another tool supporting our work, like a word processor, spellchecker, or content search engine. One exception is “mixing” our own texts. For example, if we have a 1,000-word product description, shortening it to, say, 100 words with the help of AI produces good results and significantly speeds up the work. If we have hundreds of products (e.g., screws or automotive parts) and make sure that the text contains no errors, no one will complain that such a description “has no soul.” Of course, the 4–12 principle is only an example; sometimes AI can shorten the time needed to create a text from 8 hours to 4, from 15 to 6, and so on.

Key strategies I use when working with text:

  • Thorough Research (Deep Search): 2025 is the era of deep search. I use the Deep Search feature in ChatGPT and Perplexity so that the model first analyzes hundreds of sources and only then writes the text. This eliminates “fluff” and reliance on generalities.
  • Choosing a Model: For texts that are meant to sound natural and “human,” I recommend Claude AI. It has far fewer of the irritating artifacts typical of AI than standard ChatGPT.
  • Working with a “Stream of Consciousness”: AI is excellent at organizing chaos. I often dictate loose thoughts to the model (using, for example, Google AI Studio for transcription), and then ask it to give them the structure of an article. This is the best way to overcome writer’s block.
  • Feeding It Knowledge: Instead of making AI guess, I upload my own materials, PDF files, or data from Wikipedia. This allows the model to work with specifics, and the risk of hallucinations (making up facts) nearly disappears.
  • The “Step List” Method: Instead of one long prompt, I use chains: “write an outline -> conduct research -> write the text -> analyze and improve it.” This process produces results that are a class better.

The Value of Text in the Age of AI: I believe that today, genuine creative work is shifting from simply writing paragraphs toward system design, material selection, and strategy. I treat AI as a modern version of dictation—I provide the idea and knowledge, while the machine handles the tedious editorial process.

7 Useful AI Tools

15/01/2025

The article presents seven practical AI tools that can be used in everyday work and private life. Perplexity is an advanced AI-powered search engine, more intuitive than the free version of ChatGPT, available without the need to create an account (Grok and Gemini currently offer similar functionality). Claude AI is the main alternative to ChatGPT, distinguished by a more human and natural writing style and the ability to upload files—its drawbacks are the lack of internet access and plugins. Google AI Studio (the developer version of Gemini) offers free transcription of recordings in Polish, model temperature adjustment, thorough content searches, and access to fine-tuning, although it does not support image generation or internet access. Google’s Notebook LM allows users to “talk” with their own notes, documents, PDFs, and YouTube links, and can additionally create podcasts lasting several dozen minutes in the form of a conversation between two people (currently only in English).

Gamma App enables the rapid creation of visually appealing presentations, documents, and websites—in 5 minutes, you can create something that looks like the result of many hours of work, with the option to publish or download it in PowerPoint format. Napkin.ai is a similar tool focused on generating and editing illustrations for documents, making it ideal for creating personalized graphics. There is an AI for That is a portal that helps you find a specific AI tool suited to your needs, although sometimes a regular internet search or Perplexity works better. All these tools offer unique features that streamline work—from research and transcription, through note organization, to creating presentations and illustrations. It is worth testing them yourself to find the solutions that best meet your needs, especially since most of them are available for free or offer a free trial period.

ChatGPT: how is this program built, and what functions does it have built around writing with a language model?

31/01/2025

Most content about ChatGPT is focused on the language model itself. In this text, I focus on its features from the user’s perspective. I focus on how the program and its interface (UX) have evolved, transforming into an advanced knowledge management tool.

Key system features:

  • Memory and Project Management: ChatGPT can now remember facts across threads (Memory). I use this to create “profiles” for projects—for example, by naming a specific writing style that the model can return to in a new conversation without having to teach it the basics again.
  • Canvas Interface: This is a powerful change in UX. Instead of constantly generating new versions of the text, I work in an editable window next to the chat. This allows for manual edits within AI-generated text—without copying it into external editors.
  • File Analysis and “Feeding It Knowledge”: I believe uploading PDFs is an underrated feature. It allows you to “talk” with books or legislation. As a result, even a free model can provide better answers than a paid version if we give it hard data instead of relying on its general knowledge.
  • Model Ecosystem: I show how to switch between models within a single thread. I use o1 for complex logical analysis and structuring, and 4o for generating graphics, browsing the Internet, or final stylistic polishing.

My philosophy of using AI: The key in 2025 is no longer simply “writing to a bot,” but context management. ChatGPT becomes a personalized assistant because I provide it with my own materials (files, descriptions of situations) and consciously manage what the model should remember and what it should discard.

AI for Students

30/12/2025

Artificial intelligence in university studies is much more than just ChatGPT, which can be used to generate ready-made answers—it is powerful support for the mind that handles the “grunt work,” leaving time for thinking and creativity. The key to effective research is using tools with access to reliable sources: Deep Search mode in ChatGPT and Perplexity for quickly getting oriented in a new topic, Consensus.app for searching peer-reviewed scientific publications, and Google’s NotebookLM, which works only with files provided by the user, eliminating the risk of hallucinations. AI also works well as a text editor (Claude, Bielik), a tool for creating presentations (Napkin.ai, Gamma.app, Gemini Canvas), and enables “vibecoding”—creating advanced Excel formulas or Python scripts using natural language.

The ethics of using AI is the most important consideration. Safe applications include proofreading your own text, explaining concepts, and generating work plans and research using dedicated tools, while it is unethical to generate entire chapters, thoughtlessly cite unverifiable sources, or claim as your own ideas invented by a machine. Universities increasingly require transparency—it is worth citing the use of AI in accordance with guidelines (e.g., APA) and describing it in the methodological introduction. Despite access to the latest models, you should not abandon traditional handwritten note-taking, which fosters creativity and better information processing. AI is a partner for discussion and verification, but curiosity and the ability to ask the right questions remain the most important qualities of a student.

Artificial Intelligence in Practice

31/12/2024

This article serves as an introduction to the practical use of artificial intelligence. You will find five non-obvious milestones in learning AI: mini fine-tuning to adapt the response style, creating your own knowledge bases by uploading documents (simple RAGs), using AI for summarization and transcription (Google AI Studio handles Polish exceptionally well), keeping notes with commands, and using plugins. I also present alternative tools: Claude.ai for its natural writing style, Perplexity.ai for research, Gamma.app for presentations, and Google AI Studio for transcription in Polish.

The text is already a year old, and although some of its content is essentially timeless, it needs to be refreshed.

Grok

Grok AI is an advanced artificial intelligence assistant created by xAI, offering significantly more features than the basic version available directly on Twitter. Grok’s main advantages are its low level of censorship while maintaining good response quality, fast performance, and integration with Twitter, providing access to up-to-date information. The program stands out with its Projects feature, which allows users to add multiple documents and discuss them across different threads—importantly, this feature is available even in the free version, unlike competitors such as ChatGPT. Additionally, Grok offers Canvas for editing documents, query automation through Tasks, and three search levels, including the advanced DeepSearch and DeeperSearch.

In terms of pricing, Grok offers a free version with a wide range of features, while paid plans start at $30 per month, with the most advanced Grok Super Heavy costing $300. This is more than competing solutions – ChatGPT Plus (20), GoogleAIPro (19.99), Google AI Pro (19.99), GoogleAIPro (19.99), or Claude AI ($17) – but the latest Grok 4 model performs very well in benchmark tests. A drawback is the lack of an option to preview code directly on the website, which other tools offer. Nevertheless, Grok is a valuable alternative, especially for users seeking a more open model with access to up-to-date data from Twitter and extensive project features in the free version.

Artificial Intelligence from Google

03/05/2025

This text is an overview of the latest AI tools from Google, supplemented by an analysis of the entire Google AI ecosystem, which lagged behind ChatGPT for two years. The breakthrough came thanks to three key solutions and a new strategy. NotebookLM is a unique note-taking assistant that allows you to upload up to 50 sources (documents, PDFs, links to websites, and YouTube videos with captions) and have a conversation with them, choosing which sources should be active. It offers automatic summaries, mind maps, timelines, FAQs, and a podcast feature—it generates a several-dozen-minute recording based on your notes, perfect for learning on the go. Google AI Studio is a free developer platform that enables you to test models before deployment, create audio and video transcriptions (excellent for Polish), generate videos using VEO 2, and build your own applications with “Starter apps.” Gemini is a chat service that has finally caught up with the competition—it offers Deep Research, which creates a well-considered plan and coherent responses spanning thousands of words, Audio Overview (short podcasts), integration with YouTube, Google Drive, and Google Maps, and Canvas for convenient text editing with export to Google Docs o

Free custom models available once.

Google has adopted a strategy different from OpenAI’s: instead of offering one versatile tool, it provides many specialized applications with deep integration into its own ecosystem. Gemini works faster than ChatGPT, has access to YouTube, Drive, and Maps, and excels at quick research and working with videos. ChatGPT still writes long-form texts better and creates editable graphics, has more advanced cross-thread memory, and a stronger Deep Search. Google’s drawback is the lack of convenient integration between its own tools. In practice, it is worth having both solutions—Google for research and integration with your own data, and ChatGPT for creative writing and linguistic precision.

Claude – a language model with a good style

Claude, alongside ChatGPT, Grok, and Gemini, is a major player in the world of language models. In practice, its main advantage is its appealing writing style. Other models may be better at gathering information, but when we want a text to be easy to read, it is useful to paste, for example, research from ChatGPT into Claude. Created by Anthropic (founded in 2021 by former OpenAI employees), Claude stands out primarily for its excellent configuration, which means it rarely hallucinates and is relatively safe—it is difficult to “jailbreak” it into generating dangerous content, which is crucial when creating customer service chatbots or tools made available to the public.

An interesting fact that isn’t mentioned in the linked article because I learned about it later: Google invested in Claude. On the one hand, this demonstrates the strength of the tool; on the other, it undermines the brand’s image as being “independent of the major giants.” On the third hand, the mere fact of Google’s investment does not mean a complete loss of independence.

Claude’s main advantage is its exceptional programming ability—it regularly ranks among the top performers in coding benchmarks, offers a huge context window capable of analyzing millions of tokens, and, above all, enables programs to run directly within the application. Users can create and share “artifacts”—simple programs that run in Claude—which is extremely convenient. Additionally, Claude provides access to artifacts created by other users, a creation and editing history, the ability to connect external applications via API, and a writing style selection feature conveniently located next to the message window.

A drawback of Claude is its relatively small number of features compared with the competition—it cannot generate images (which is standard in other models), and internet access was introduced relatively late. In the paid version, Claude can be connected to other applications, while downloading the desktop app provides access to computer files and the ability to control other programs (an experimental feature). Claude Code is also available—a tool that integrates the model with a programming environment, eliminating the need to copy code manually. Anthropic has adopted the philosophy of “constitutional AI,” with built-in ethical principles, and regularly updates the model—in May 2025, it introduced a new API for developers that enables them to build AI agents for automating complex tasks. Using Claude also provides a safeguard in case ChatGPT fails and an opportunity to gain a better understanding of different approaches to creating language models.