What is generative AI?

It is 11:40 at night. Meera’s cousin Kabir turns 21 tomorrow, and the family WhatsApp group expects a birthday shayari from her. She types into a chat app: “Write a funny four-line Hindi shayari for my cousin who sleeps till noon and loves cricket.” Five seconds later, a brand-new poem appears. Nobody wrote it before. Nobody saved it anywhere. The app made it up, right then, for Kabir.

That feeling of “wait, where did this come from?” is exactly what this lesson answers. By the end, you will know what generative AI is, how it differs from the AI that was already inside your phone, and why it suddenly got so good.

Picture a big Indian wedding. Two people are working hard there. The first is the catering manager at the entrance. Each guest walks up, and the manager points: veg counter left, non-veg counter right, kids’ corner straight ahead. Useful work, but the manager never creates anything. Every guest is sorted into a box that already exists.

The second is the mehendi artist. A guest says, “Something with peacocks, and hide my fiancé’s initials in it.” The artist has drawn thousands of designs before, so they know what peacocks, vines and paisleys look like. But the design on this hand is new. It did not exist ten minutes ago.

For years, most AI was the catering manager. It looked at something and sorted it or predicted a number: is this email spam, is this payment fraud, how long will this cab ride take. Generative AI is the mehendi artist. It looks at what you asked for and produces something new.

Here is a quick test you can use for any app. Ask: is the answer picked from a fixed set of options, or is it a brand-new thing? “Spam” or “not spam” is a fixed set. A reply to your landlord that did not exist a second ago is new. That second kind is generative.

Think about an auto-rickshaw meter. Someone wrote its rule: a fixed fare for the first stretch, then so many rupees for every extra kilometre. The meter follows that rule perfectly and can do nothing else. Ask it to write a poem and it will stare back at you.

Now think about your mother’s cooking. Ask how much salt goes in the dal and she says, “Andaaz se.” By feel. There is no rule written anywhere. She has cooked dal thousands of times, tasted it, adjusted it, and her hands learned the pattern. She can even make a dish she has never cooked before, because the patterns carry over.

Generative AI works much more like the andaaz than like the meter. Nobody writes rules such as “a birthday shayari must rhyme on line two”. Instead, the system studies an enormous number of examples and slowly picks up the patterns by itself. Lesson 3 shows that learning loop step by step.

Text gets the most attention because chat apps like ChatGPT or Gemini are everywhere. But the same idea, learn patterns and then create, works for many kinds of content.

Some tools now handle several kinds at once. You can show a photo of your fridge and ask for a dinner recipe, or speak a question and hear the answer. You will meet these “multimodal” tools later in the course.

Researchers have tried to make computers write and draw for decades. The early results were clumsy. Three things came together in the last ten years or so, and the difference became hard to ignore.

• Lots of examples: the internet gave researchers a huge amount of text, code and images to learn from.
• Powerful chips: graphics processors, the chips first built for video games, turned out to be great at the maths of training.
• A better design: in 2017, researchers at Google published the “transformer”, a model design that reads a whole passage at once and learns which words relate. You will meet it properly in module 2.

When ChatGPT launched in late 2022, millions of people tried a generative AI tool for the first time within weeks. Since then, these tools have moved into email apps, office software, phones and coding editors. That is why “GenAI” now comes up in classrooms, placement talks and office meetings alike.