GenAI for Dummies

This article talks about what GenAI is, how it changed the IT industry, why it has become an essential skill for everyone today, and how it’s going to change the world in the coming days. You can also read real-life scenarios which can easily be implemented by GenAI applications. Let’s do a deep dive into it!

Introduction

GenAI is the acronym for Generative Artificial Intelligence. Generative, because it generates new data as the output of its learning. If the above language sounds tough to you, then you can imagine GenAI as a movie’s actor who can play any role which you want it to be. Be it a personal assistant or translator or proofreader or business analyst or developer or anyone. There are many generative AI models readily available at present such as OpenAI’s GPT & DALL-E, Google’s Gemini, Facebook’s Llama, Databricks DbRx, and so on.

TradAI vs GenAI

Traditional AI (TradAI) is a classical AI that involves rule-based systems, or machine learning or other techniques like supervised learning that target a specific problem or task. This is very effective when we try to solve specific, well-defined problems. The limitation with traditional AI approach is it requires a significant number of labeled data. It also struggles with new content generation problem statements and in creativity as well.

On the other hand, generative AI (GenAI) creates models that generate new content that could be text, image, videos, music, and many more. It learns patterns and structures within the data to produce output which is not known beforehand. It is perfect for most of the real-life scenarios which we deal with in our day-to-day life. It often involves unlabeled data for its learning.

Superstar Entry – The GPT, ChatGPT!

The month was November, in the year 2022, and the date was 30th. At that time, an unknown small startup company with the name OpenAI suddenly became the talk of the town due to the product it launched on the internet with the name ‘ChatGPT’. Media houses, social network sites, and all the other platforms just couldn’t stop themselves from praising this product as it became viral across the globe. That’s the first time people experienced the power of Artificial Intelligence in real-life scenarios. There were many new terms which became popular in the following year. Terms like GenAI, LLM, RAG, RAGA, fine-tuning, etc. suddenly become the buzzwords since 2023 and still are.

ChatGPT Impact

This one GenAI-powered application revolutionizes the data industry left, right, and center. The impact of ChatGPT was so huge that all IT giants fell into the race of who releases the best large language foundational model (LLM) for text, images, and videos. As of 2024, there are approximately more than 300 LLMs available worldwide and mostly from the USA and China (80% of global LLMs). Some of the big names which are actively releasing LLMs are Microsoft, Google, Nvidia, Facebook, and OpenAI. There is not a single industry which is untouched by this advancement of AI and data science. Finance, healthcare, e-commerce, content moderation, hospitality, are few to mention out of many.

Post-ChatGPT Era

It is interesting to know what all advancements happened after the ChatGPT revolution in the GenAI world. NLP-based models are the AI models that understand English in a way like any other English scholar or sometimes even better than them. Similarly, they are intelligent enough to understand many other naturally spoken world languages like Hindi, Spanish, Mandarin, and many more. GenAI models are the further enhancement of the NLP models that are better and more context savvy.

As mentioned in the history section above, it all begins at the end of the year 2022. The term tossed was the foundational large language model (LLM). LLM is the first thing that appeared in the modern AI revolution called GenAI. These are GPT, Llama, Mistral AI, Gemini, and many more. Some of them were open source while others are paid. These LLMs are the trained model from the very large internet data that can perform numerous tasks. These tasks are text summarization, text completion, story generation, question answering, etc. Though LLM serves many purposes, they are more generic in nature and can’t handle specific use cases. This creates a void and requires someone to fill this.

To overcome the above problem, a new technique was widely accepted, which has opened the door for LLMs to output more refined and domain-specific generative data. The term used for this technique in the GenAI world was fine-tuning. Fine-tuning a base model gave that extra edge to LLMs by which the derived models can generate the answers more accurately when discussed for specific domains simply by relearning on domain-specific data. This technique is good to resolve the issue of specialization problem in LLMs but still, it suffers from the stale learning data issue. It lacks data freshness and has always been fine-tuned on the backdated labeled dataset and can’t respond appropriately to the prompts involving the latest information.

A Retrieval Augmentation Generation (RAG) technique was introduced later to solve the data freshness issue. RAG works with the latest data and involves prompt engineering at multiple levels. Further agents have been introduced that talk to these prompts at different levels to accomplish a certain action. And this evolution continues.

GenAI Real Life Use Cases

There are many real-life use cases which you think can easily be solved by using the GenAI techniques. Some of these are:

1. Medical insurance companies’ interactions are always a pain, and sometimes you need to wait for almost 1 hour to resolve the small dispute in the bills. GenAI can reduce this latency from hours to minutes or seconds sometimes.
2. FinTech business is highly data-driven. The existing data can be leveraged to come up with GenAI apps that effectively reduce many repayment and other critical issues which the company and customer both see on a day-to-day basis.
3. Real estate business is also a great place that can improve abnormalities in a big way if they go with the latest AI technology.
4. Customer call support division has a lot of issues with dummy bots and lack of proficiency over call. This can easily be solved with generative AI-based applications.

Challenges

GenAI-based applications are not perfect and it also suffers from its own challenges. Issues like data hallucination and personal data protection are some of the many problems it has. Hallucination is more towards propagating false information with full confidence that looks like a true response. And data protection is more to do with the PII data exposure. People are working towards it big time to reduce the appearance of these issues to the maximum extent. But nothing is perfect in the world and every blessing comes with some unpredicted curse.

Summary

We have seen what the GenAI term means and how it evolved over the period, its popularity and significance in the world today. This article laid a strong foundation for all non-technical users who keep wondering what GenAI is and why it is important today so that you understand the key happenings in the AI world. Here on, you can further leverage your knowledge on GenAI by going much deeper into it via advanced resources. Thank you and Happy Human Learning!

About the Author

Ashish Garg is a GenAI Data Scientist and Databricks Engineering Architect at HYR GLOBAL SOURCE Inc. With his extensive expertise in Generative AI and Databricks, Ashish has been instrumental in delivering cutting-edge solutions to various esteemed clients. His innovative approach and deep understanding of the latest technologies have made significant impacts in the field of AI and data science.

Ashish Garg