Generative AI

As the name suggests Generative AI is a type of AI that generates new contents unlike other types that provides output based on existing data and perform actions like classification or identification within available set of data. Generative AI is specifically designed to generate new content as its primary output whether it’s text, images, audio or anything else.

In this article I am covering only Generative AI, I will cover other types later. Also have a look here for my article covering What is AI, ML and Microsoft AI Builder.

How generative AI works

What is behind? How Generative AI come to life? To understand let’s have an example, if you are standing in front of a table on which multiple bottles/pots are placed and someone asked you to hand over sugar pot you can easily identify which one is it and can hand it over to the person, how? Because you or your mind is trained since ages with multiple bottles/pots and sugar, so you can easily identify the desired object, same is the case with Gen AI models. Models are trained with millions of records and patterns to generate specific results based on given input. With the help of knowledge (data) and logic (coded in algorithms) a model understands how to respond.

Neural networks, which mimic the human brain, enable generative AI to create original and meaningful content by learning from the data’s patterns and relationships. Generative AI models can train themselves using different learning approaches, such as unsupervised or semi-supervised learning.

Generative AI Models:

Various Generative AI models are developed by companies like OpenAI or Google etc., these models are getting mature over the period of time with the usage and generated results. As these models needs a lot of expertise and huge investments various big companies or universities are working to develop them and either releasing them as an open source or paid service to utilize in real scenarios.

we can get benefit from these models depending on the expertise or need that we have. For example:

• For beginners: it can offer guidance, feedback, and suggestions to improve their skills and knowledge. They can also generate examples, exercises and quizzes to test their understanding and progress.
• For professionals: it can assist them with complex and creative tasks, like generating code, reports, presentations etc. They can also automate their routine work like data analysis, translation, etc.
• For hobbyists: it can help to explore their interests like generating art, stories, games etc. Can also inspire them with new ideas & styles.

There could be many other use cases of generative AI models across different industries like Marketing (content), Gene sequencing, develop code, generate articles, generate videos, 3D models, design, Image, generate voice etc.

Creating Content

One can start with ChatGPT by giving instructions to write an essay on any selected topic, can be done either on ChatGPT or Bing Chat OR one can create pictures OR videos by giving text instructions to related available models OR if you have expertise in technology you can go to GitHub and choose available generative AI model as per need and can get the desired task done.

In case if you are a programmer, you can also write your own notebook by taking the model from GitHub. A notebook is a tool for writing and running the code which also offers various settings or options to get a tailored outcome from the model however a model is set of algorithms trained on specific data.

Well known Generative AI Models

Here are some well-known generative AI models:

Large language models (LLMs)?are a kind of machine learning model that are trained on massive amount of data to realize and generate natural language. They are skilled at performing a broad spectrum of tasks such as recognizing, summarizing, translating, predicting and generating text. LLMs are artificial neural networks, predominantly transformers and are trained using self-supervised learning and semi-supervised learning. They operate by taking an input text and iteratively predicting the next token or word. Prominent examples of LLMs include OpenAI’s GPT models, Google’s PaLM & LaMDA models, Meta’s LLaMA models and Hugging Face BLOOM.

Few famous Chat Bots based on LLMs are ChatGPT by OpenAI, Bard by Google, Bing by Microsoft etc.

GitHub Copilot3: It is a computer program that uses AI to help programmers make better code quicker. It learns from a lot of code that other people have shared and can give ideas for whole functions or classes based on a few words. It’s also a type of LLM.

Midjourney3: It is a text-to-image product that generates images from textual descriptions. It is trained on a large dataset of images and their textual descriptions. The better you describe the better you get the results.

Few famous Image generation software are: DALL-E, Midjourney, Stable Diffusion etc.

Runway Gen-23: It is a text-to-video product that generates videos from textual descriptions. It is trained on a large dataset of videos and their textual descriptions.

Generative Adversarial Network (GAN) is a?deep-learning-based generative model that consists of two neural networks: the generator and the discriminator. The generator produces fake data and the discriminator distinguishes it from real data. The two networks compete against each other in a zero-sum game, where one agent’s gain is another agent’s loss.?GANs are used for various applications such as image generation, text generation, and natural language processing.

Variational Autoencoder (VAE) is a type of neural network that can be used for anomaly detection. It is a generative model that learns to encode data into the latent space and subsequently decode it. We detect anomalies by measuring the dissimilarity between the original input and the reconstructed output. If the reconstruction deviates significantly from the input, it indicates an anomaly. One example is GitHub, another is TensorFlow available on Analytics Vidhya.

Consideration

Results generated by generative AI models may encounter difficulties or may give wrong or biased results such as factual errors, language bias, gender bias, racial bias, and political bias. One should always check the generated results before sharing it further to avoid any issue.

The reason behind these possible errors is that these models are trained on large amount of data and they work on pattern so they suggest based on the pattern they learned which could be incorrect. There is a term hallucination in generative AI where AI basically makes up its answers that sounds right but are not even at times if provides references that doesn’t exists at all. This is because Gen AI is very good at patterns and it creates results that are statistically likely but it’s not necessary that they are true so it’s always good to fact check before moving forward. ?

Gen AI Prompts

A Gen AI prompt is a way of instructing an artificial intelligence system to generate a specific type of output, such as text, image, video, or code. A gen AI prompt can be a word, a phrase, a question, or an example that helps the AI understand what you want it to create. If you want the AI to write a poem, you could give it a prompt like this: “poem, romantic, 12 lines, roses, moonlight”. Based on this prompt, the AI will generate some output, as Gen AI generates it could be different every time and different models could respond differently. A prompt can help you unleash your creativity and explore new possibilities with the help of AI. You can also customize your prompts according to your preferences, such as genre, tone, style, length, or keywords. The more specific and clear your AI prompt is, the more likely the AI will generate a relevant and coherent output.

This is it for now, next topic related to AI could be Prompt Engineering.