Generative AI for Beginners on Basic Concepts - Short Narrative

What is Generative Artificial Intelligence?

There are two terms, one is Artificial Intelligence and the other is Generative.

So, Artificial Intelligence (AI) is a fancy term for saying we get a computer program to do the job that a human would otherwise do. Any Generative is a bit into it, i.e., we are creating new content that the computer has not necessary seen it, it has seen parts of it and it is able to synthesize it and give us new things.

So, what would this new content be?

It could be audio, it could be computer code so that it writes a program for us, it could be a new image, it could be a text like an email or an essay, so basically the Large Language Models make predictions what the continuation of the text will be and generate creative new contents.

So, basically Language Models make prediction what the continuation of the text will be.

What has changed?

We have Google Translate, Siri on iPhone, Amazon Alexa, then when you write text on your mobile or even when you want to search in Google, i.e., Auto Completion, and these are all some form of AI. They are simple purpose systems. So, what have changed and we will start with that!!

You can use ChatGPT to write text for you, for example you have prompt and a potential prompt would be “I am writing an essay about the use of mobile phones during driving. Can you give me three arguments in favor? This is quite sophisticated, Or act as a JavaScript Developer, write a program that checks the information on a form. Name and Email Ids are required but address and age are not.

What is the technology behind it?

The technology that it turns out is not entirely new or extremely innovative or extremely difficult to comprehend. So, all these GPT model variants, I am just using GPT as an example because public knows a lot of these via articles etc. but there are other models that is also used but they all ?work in one principle which is called Language Modelling.

What does the Language Modelling do?

The task of the Language Model is that we have a sequence of words and there is a context available in that sequence of words and the Language Model will predict what comes next.

Say for e.g.,

Assuming my context is the phase “I want to”, the Language Modelling tool will predict what comes next. Now there can be several predictions like the following :

• I want to shovel
• I want to play
• I want to swim
• I want to eat

And depending on what we choose, whether it’s shovel or play or swim, there is more continuation.

So, for shove, it will be snow, for play it can be tennis or video, swim doesn’t have a continuation and for eat it will be lots and fruits.

In real like, the Language Model has seen a lot of text and it knows what words follow which other words. We used to count these things, so I would download a lot of data and I would count “I want to shovel” and how many times does it appear and what are the continuations

But now we use Neural Networks that don’t exactly count but predict, learn in a more sophisticated manner.

So, ChatGPT and GPT variants are based on the principle of I have some context and I will predict what comes next. The task of the Language model is there is a context and they predict what comes next.

#HappyReadingOnGenerativeAI