GenAI - Ramworld

Generative AI

We live in the age of technology and development, and the technological revolution, including artificial intelligence. It’s a controversial subject with different opinions, from supporters and opponents, fearing that he would control human life and put an end to many jobs, believing that he might be able to replace them. But artificial intelligence will not replace any function in the world, it is a tool dedicated only to performing certain tasks in a particular way.

There’s a saying that artificial intelligence can’t take the jobs of people who know how to use it.

The meaning of this sentence is that artificial intelligence education and awareness-raising are essential so that jobs are not lost and we start blaming it. Artificial intelligence has facilitated a lot of things in our daily and practical lives, making use of that in our development and making use of the services provided.

Today we talked about another kind of artificial intelligence, and I decided not to invent the title of the article so that I wouldn’t get out of context, that is, obstetrical artificial intelligence.

Generative AI is one of the branches of artificial intelligence that can create different kinds of visual or written content based on human input.

How can you create content only from a sentence? That’s what we’re going to talk about in this article.

First of all, what is artificial intelligence? And what is robotic learning?
Let’s start with a quick review of the concept of artificial intelligence and its branches.

Artificial intelligence is a branch of the computer, and automated learning is a branch of artificial intelligence.

Artificial intelligence is a set of orders that create smart systems that can think and act like humans, and auto-learning is a program that trains a model based on input data.


An application for monitoring health and fitness. Artificial intelligence is an application by analyzes personal data such as age, state of health, and the target to be reached. It then establishes a health system and proposes it to you if you want to comply with it.

Automated learning then collects user data and learns from the data so that it is developed from itself to provide more appropriate systems that are adapted to the user ' s personal needs. Automated learning: two categories of learning models: supervised learning and unsupervised learning. Supervised learning can be defined by two signs, either classified or not classified.

The monitored learning data, which are classified, will contain markings such as name, type, or even number, while the unclassified data do not contain markings with which they can be identified.

Like the image recognition system, we were supposed to build a system to identify images classified as car type. We collect the images and their types and then we start training the model in the images.

Unsupervised learning is all that has to do with data discovery, so looking for raw data and then sorting it out so we can see if it falls into groups naturally. Like collecting news in the world based on topics without identifying more important topics than others.

We sail to get to deep learning, a branch of robotic learning, but in more depth.

Deep learning uses artificial neural networks that allow it to address more complex patterns. And here’s the difference between deep learning and mechanical learning, deep learning can address more complex patterns than mechanical learning.

The concept of artificial neural network.
A network of knots or, as we call them, neurons, interconnected. The connection of artificial neurons enables them to perform tasks through data processing and forecasting.

Deep learning models typically contain many layers of neurons that enable them to perform more difficult tasks and learn more complex patterns than automated learning models.

Synthetic networks can use disaggregated and unclassified data, which we call semi-supervised models.

Similar forms are supervised and help train the neural network in a small amount of disaggregated data and a large amount of unclassified data.

Let’s take an example, filter e-mails. The semi-supervised model trains the neural network using several classified (undesirable) e-mail messages and then uses these data to analyze the rest of the messages whether or not they correspond to the unwanted message pattern. Disaggregated data helps the neural network to learn basic mission concepts, while non-disaggregated data helps the neural network to learn new examples.

Generative artificial intelligence and deep learning.

We get to the theme of the article, that is obstetrical artificial intelligence. We talked at the beginning that A.I. is one of the branches of A.I.A., and it’s also a sub-part of deep learning. Generative artificial intelligence can create different kinds of visual or written content based on human input. This means that obstetric intelligence also uses artificial neural networks and processes disaggregated and ungraded data using the three models, the supervised and unsupervised model, and finally the semi-supervisory model.

Deep learning models, or mechanical learning models in general, can be divided into two types:

It’s discriminatory and it’s generational.

The discriminatory model is one of the models used to classify data, expect to name certain data points, and identify the relationship between the characteristics of the data points and the labels. The generational model creates new data based on the likelihood that it learns about data already entered.

How’s the A.I. process going?

Let’s say we’re now at the ChatGPT site and we’ve got him a question or a mission that’s creating an XO game. Just what we got him on the job, he listed the solution, either he worked or he made some minor mistakes. And that’s not a flaw, because it’s not based on being programmed, but rather on being a tool that helps to program easily, simply, and fast. But, how did that happen?

When OpenAI developed ChatGPT, it developed and trained it in several modules of the software languages by collecting, training, and revising the Python-language projects to the test and verification stage.

Secondly, it analyzed the requirements in several stages. The first stage is the creation of a game and then the second stage is the name of the game and the last stage is the language used to create it.

Thirdly, based on his training modules, he collected information and software lines from online projects and built the game on this knowledge.

Fourth, he built the game based on the logical analysis of the game, for example, the game will have either one winner or no winner. If the winning conditions are met, the winner will be determined.

Fifth, he documented the user’s computer lines after he intended to test them.

We get to the most important question, how do we know if the result or the output belongs to the AI or not?
It’s generational intelligence if it’s text, visual, or audio, and non-generative if it’s a number, class, or probability. There are several models of reproductive artificial intelligence:

Image Generation Model: The model takes input either as a picture or text, and translates it into output either as a text describing the image or as a new image similar to that initially introduced.

Text Generating Model: The model takes the text that has been introduced, translates it, and analyses it to be either a new text, images, or a decision to recommend it.

The model of text generation learns about language patterns through training data, and then, in the light of some texts, predicts what will come next.

Of course, artificial intelligence is a software order written by a programmer, and no matter how much computer learning and deep learning are used, there can be mistakes or outputs that are irrational, which we call hallucinations.

They are words or phrases that are created by the model and are often irrational or grammatically incorrect. This is due, inter alia, to:

The model was not trained in sufficient data
Training of the model in relatively incorrect data, or as we call it (unprocessed data)
Not giving the form enough time to rehearse the input data.
Large language models:
A sub-set of deep learning is a tool of reproductive artificial intelligence. Large linguistic models and generational artificial intelligence are part of deep learning and work together effectively. Generative artificial intelligence is trained using large language models that are also pre-trained and modified so that they can perform their specific tasks. There are three main features of large language models:

Size: refers to the quantity of data and the large numbers of data it possesses.
Public: All human languages include the ability to communicate and deal with a variety of tasks.
Pre-training: These modules have been pre-trained and their performance continuously improved.
One of its advantages is:

One model can be used for several tasks.
Increase and improve the model ' s performance based on the amount of data input.
These include:

Predict the next word.
Response to specific input instructions
Dialogue through the expectation of the next response.
In conclusion, artificial intelligence is a sea of information and continuity that is required in this age, and let us continue this path and be leaders of development and digital transformation. But let us not forget that much of the development is to make society aware of the risks and technical and artificial intelligence.


Rousol Jamal