{"id":7103,"date":"2025-08-02T13:31:20","date_gmt":"2025-08-02T10:31:20","guid":{"rendered":"https:\/\/store.nit.bg\/the-chat-bot-that-doesnt-think\/"},"modified":"2026-08-02T10:53:23","modified_gmt":"2026-08-02T07:53:23","slug":"the-chat-bot-that-doesnt-think","status":"publish","type":"post","link":"https:\/\/store.nit.bg\/en\/the-chat-bot-that-doesnt-think\/","title":{"rendered":"The chat-bot that doesn’t think"},"content":{"rendered":"\n
\n\n
\n\nMost people believe that ChatGPT is the artificial intelligence. They are wrong. Calling ChatGPT “artificial intelligence” is like calling the microwave “cooking” – it’s a single appliance, not the entire kitchen. \r\n\r\nChatGPT really took the world by storm for a few years, which is why many now consider chatbots to be artificial intelligence itself. In fact, it is just one particular tool built on a single technique – nothing more. \r\n

1\ufe0f\u20e3 What is “artificial intelligence” (AI)?<\/h3>\r\n
Machine Learning (ML)<\/strong><\/summary>Learning from data without writing rules for it. Example. \r\n\r\n<\/details>
Natural Language Processing (NLP)<\/strong><\/summary>Understands and writes human language. Example. \r\n\r\n<\/details>
Computer Vision (CV)<\/strong><\/summary>“Sees” and analyzes images. Example: Face ID unlocks you even with a hat and glasses because it has studied millions of faces from different angles. \r\n\r\n<\/details>
Robotics<\/strong><\/summary>Moves and works in the real world. Example: the robot in the Amazon warehouse picks up boxes with millimetre precision and doesn’t get tired. \r\n\r\n<\/details>
Expert systems<\/strong><\/summary>Apply ready-made rules. Example: software in a hospital inputs symptoms and outputs the most likely diagnosis according to medical protocols. \r\n\r\n<\/details>\r\n

2\ufe0f\u20e3 From AI to ChatGPT – a 7-step “ladder descent”<\/h3>\r\n
    \r\n \t
  1. Artificial Intelligence (AI)<\/strong>\r\nthe whole science, from a chess computer to a self-driving drone.<\/li>\r\n \t
  2. Neural Networks (NN)<\/strong>\r\nmathematical “neurons” that together capture complex shapes such as a face or handwriting.<\/li>\r\n \t
  3. Deep Learning (DL)<\/strong>\r\nhundreds of layers of neurons – the more, the finer the details it recognizes.<\/li>\r\n \t
  4. Transformers (2017)<\/strong>\r\narchitecture that reads an entire sentence at once and connects a word from the beginning with one from the end.<\/li>\r\n \t
  5. Large Language Models (LLMs)<\/strong>\r\nbillions of parameters “read” the entire Internet; GPT-4 is an example.<\/li>\r\n \t
  6. GPT-4<\/strong>\r\nthe exact engine hidden behind ChatGPT – more powerful, but without the chat-wrapper.<\/li>\r\n \t
  7. ChatGPT<\/strong>\r\nnice chat window, added conversation memory and safety filters.<\/li>\r\n<\/ol>\r\n

    3\ufe0f\u20e3 What happens “under the hood” of ChatGPT?<\/h3>\r\n
    Learning as a child – two phases<\/strong><\/summary>Phase 1:<\/strong> Reads the whole internet without labels – learns grammar, facts and style.\r\nPhase 2:<\/strong> People show good and bad answers – model adjusts to be helpful and polite.\r\n\r\n<\/details>
    Transformer in 3 sentences<\/strong><\/summary>1) Divides text into small chunks (tokens).\r\n2) Each chunk “looks at” all the others (self-atttention).\r\n3) Adds position codes to know the order.\r\n\r\n<\/details>
    Generating the response<\/strong><\/summary>Does not search a database but predicts the next word word by word. If there is no information – makes it up in a confident tone. \r\n\r\n<\/details>\r\n

    4\ufe0f\u20e3 Issues everyone should know about<\/h3>\r\n
    Hallucinations<\/strong><\/summary>He makes up scientific papers and cites them with DOIs. Always check a second opinion. \r\n\r\n<\/details>
    Loss of context<\/strong><\/summary>After a few paragraphs I forget the beginning. Break long tasks into parts. \r\n\r\n<\/details>
    Fixed context<\/strong><\/summary>GPT-4o remembers ~300 pages. Don’t hand it a whole book at once. \r\n\r\n<\/details>
    Repeatability<\/strong><\/summary>Often chooses the same “random” number. Raise the temperature for variety. \r\n\r\n<\/details>\r\n

    5\ufe0f\u20e3 How to Use ChatGPT Smartly – 5 Quick Rules<\/h3>\r\n
      \r\n \t
    1. Verify<\/strong> what is important in a reliable source – confidence does not mean truth.<\/li>\r\n \t
    2. Break<\/strong> complex tasks into steps – the chat-bot loses thread on large blocks.<\/li>\r\n \t
    3. Give context last<\/strong> – that’s what you remember best.<\/li>\r\n \t
    4. Use for drafts<\/strong> – translation, ideas, structure; then edit.<\/li>\r\n \t
    5. Don’t rely on critical topics<\/strong> – law, medicine, finance – without human verification.<\/li>\r\n<\/ol>\r\n

      6\ufe0f\u20e3 Advanced mini-dictionary<\/h3>\r\n
      \r\n
      AI<\/strong>\r\nA big umbrella over all smart machines, from YouTube recommendations to driverless cars.<\/div>\r\n
      AGI<\/strong>\r\nThe Hollywood assistant who can cook, heal and go out for coffee – more fiction.<\/div>\r\n
      ML<\/strong>\r\nThe computer looks at examples and learns the rules itself – without a human typing “if-then”.<\/div>\r\n
      DL<\/strong>\r\nHundreds of layers of artificial neurons – the more the finer the detail it picks up.<\/div>\r\n
      NLP<\/strong>\r\nFrom “translates word for word” to “understands irony and context”, this is NLP.<\/div>\r\n
      CV<\/strong>\r\nTurns pixels into sense: face, car, cat, tumor on x-ray.<\/div>\r\n
      LLM<\/strong>\r\nGPT-4 without the chat wrapper – the brain that never sleeps and has read the entire Internet.<\/div>\r\n
      GPT<\/strong>\r\nThe exact model of OpenAI – the engine that powers ChatGPT.<\/div>\r\n
      RLHF<\/strong>\r\nPeople appreciate the responses; the model is being adjusted to be more polite and helpful.<\/div>\r\n
      RAG<\/strong>\r\nFirst search documents, then write – reduce fabrications.<\/div>\r\n
      GAN<\/strong>\r\nOne AI paints a fake picture, the other catches it – after 1000 rounds they become almost real.<\/div>\r\n
      CNN<\/strong>\r\nFilters that look for edges, shapes, eyes – so your phone recognizes your face.<\/div>\r\n
      RNN\/LSTM<\/strong>\r\nRemember the order of words or notes – useful for weather forecasting or music.<\/div>\r\n
      CoT<\/strong>\r\nMakes the AI think out loud, “First… then… therefore…” – more precise answers.<\/div>\r\n
      VC<\/strong>\r\nYou speak to the computer in English “make me a button” – it writes the code.<\/div>\r\n
      ASR<\/strong>\r\nAs Siri translates “What is the meaning of life?” into letters.<\/div>\r\n
      SOTA<\/strong>\r\nEvery time someone improves the accuracy by 0.5%, it becomes the new SOTA.<\/div>\r\n
      FLOPs<\/strong>\r\nThe more FLOPs, the faster the model is trained or responds.<\/div>\r\n<\/div>\r\n

      7\ufe0f\u20e3 3 sentences to remember<\/h3>\r\nChatGPT is like a microwave for words<\/strong>: fast and convenient, but you can’t cook everything in it.\r\n\r\nAI is the whole kitchen<\/strong> – ovens, hobs, recipes, cooks.\r\n\r\nUse it in moderation, check the results<\/strong> and never forget that behind every “smart” sentence there are only numbers and probabilities.\r\n\r\nIf you feel like going down from theory to practice, store.nit.bg<\/a><\/strong> is waiting for you with ready-made, practical online training on the subject – from scratch to daily use. There, you’ll learn how to implement ChatGPT, Microsoft Copilot and Google Gemini into your workday without being a programmer, and you’ll learn the basics of artificial intelligence and machine learning adapted to reality. \r\n\r\nThe content of this article is based on Dr. Milan Milanovic’s original publication “ChatGPT is not AI<\/a>“.<\/em><\/span>\n\n<\/div>\n\n<\/div>\n","protected":false},"excerpt":{"rendered":"

      Most people believe that ChatGPT is the artificial intelligence. They are wrong. Calling ChatGPT “artificial intelligence” is like calling the microwave “cooking” – it’s a single appliance, not the entire kitchen. ChatGPT really took the world by storm for a few years, which is why many now consider chatbots to be artificial intelligence itself. In […]<\/p>\n","protected":false},"author":1,"featured_media":7102,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[165,164],"tags":[],"class_list":["post-7103","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-information-technologies","category-informaczionni-tehnologii"],"blocksy_meta":{"styles_descriptor":{"styles":{"desktop":"","tablet":"","mobile":""},"google_fonts":[],"version":7}},"_links":{"self":[{"href":"https:\/\/store.nit.bg\/en\/wp-json\/wp\/v2\/posts\/7103","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/store.nit.bg\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/store.nit.bg\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/store.nit.bg\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/store.nit.bg\/en\/wp-json\/wp\/v2\/comments?post=7103"}],"version-history":[{"count":9,"href":"https:\/\/store.nit.bg\/en\/wp-json\/wp\/v2\/posts\/7103\/revisions"}],"predecessor-version":[{"id":8302,"href":"https:\/\/store.nit.bg\/en\/wp-json\/wp\/v2\/posts\/7103\/revisions\/8302"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/store.nit.bg\/en\/wp-json\/wp\/v2\/media\/7102"}],"wp:attachment":[{"href":"https:\/\/store.nit.bg\/en\/wp-json\/wp\/v2\/media?parent=7103"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/store.nit.bg\/en\/wp-json\/wp\/v2\/categories?post=7103"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/store.nit.bg\/en\/wp-json\/wp\/v2\/tags?post=7103"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}