Models trained on massive amounts of human text to understand and generate language. They can write, summarize, translate, and answer questions in a natural style.
ChatGPT
Claude
Gemini
Copilot
What makes ChatGPT write, Midjourney draw, and ElevenLabs speak? In this lesson, we decode GPT and learn about the main fields where Generative AI operates.
Generative AI is the type capable of producing new content — text, images, audio, video — instead of just classifying or analyzing. It does not retrieve stored answers, but rather synthesizes new ones each time.
Traditional AI answers: "This email is spam." — Generative AI writes: "Dear John, I am writing to you regarding..."
GPT is an acronym for three words that describe how the model is built and how it works:
The model produces new content — text, image, code — that did not exist before. It synthesizes its answer instead of retrieving it.
Before you use it, the model was trained on massive amounts of text from the internet and books — hundreds of billions of words.
The technical architecture that makes the model capable of understanding the context and relationships between words regardless of their distance in the sentence.
GPT is the model developed by OpenAI that powers ChatGPT. But there are other models that follow the same concept: Gemini by Google, Claude by Anthropic, and Llama by Meta — all of which are large language models that differ in details, not in principle.
When you write a question to ChatGPT, this is what happens behind the scenes:
The text you enter is called the Prompt
It analyzes each word and its relationship to what comes before and after it
It chooses the most likely word based on what it learned
Word by word until the answer is complete
The model predicts the most likely words — it does not search a database of facts. Therefore, it might make mistakes or invent information that sounds logical. Always verify important information.
Generative AI does not operate in a single field — there are four main fields, each specializing in a different type of data:
Models trained on massive amounts of human text to understand and generate language. They can write, summarize, translate, and answer questions in a natural style.
Models that deal with images and video — they can generate images from text, recognize faces, read documents, and analyze video clips.
A field concerned with understanding and analyzing human language — sentiment analysis, information extraction, topic identification, and answering questions from within texts.
The complementary field to NLP — instead of understanding text, it generates new text. It converts data, numbers, and events into readable sentences and articles.
These two terms are often confused — here is the difference with one example:
Reads existing text and analyzes it — extracts meaning, sentiment, or information from it. The input is text, and the output is understanding or classification.
Generates new text from data or instructions — converts numbers and data into readable language. The input is data or a request, and the output is text.
Tools like ChatGPT and Copilot combine NLP and NLG at the same time — they understand what you write (NLP) and generate a response to it (NLG). The difference is theoretically important, but practically they are two sides of the same coin.
| Application | Field | What exactly does it do? |
|---|---|---|
| Google Translate / DeepL | NLP NLG | Understands the source language and generates a natural translation |
| Spam Filter in Gmail | NLP | Reads the email content and classifies it: important or spam |
| Microsoft Copilot / ChatGPT | LLM | Answers questions, writes, and summarizes in natural language |
| Face ID on Mobile | Computer Vision | Analyzes your face image and compares it with the saved model |
| Auto-captioning on YouTube | NLP NLG | Understands spoken speech and converts it into written text |
| DALL·E / Midjourney | Computer Vision | Generates new images based on a text description |
| Automated BI Reports | NLG | Converts dashboard numbers into a written summary |