Right Now ChatGPT is one of the best Chat bot in the world. so now we are going to Discuss the Behind the Scenes of ChatGPT.
Table of Contents
Understanding Basic Terminologies
What is AI?
Before diving into ChatGPT, let’s take a small detour to understand some basic terminologies which are very important. What is AI? Now, I know you have heard it a lot, so let me explain it to someone who still confuses it. AI is Artificial Intelligence. Under Artificial Intelligence, there is a subset called Machine Learning. Within Machine Learning, you have Deep Learning. So, all this falls under Artificial Intelligence. In very layman terms, Artificial Intelligence is a process of taking human intelligence and putting it into a machine. You want a machine to work like humans and mimic the things we do on a day-to-day basis and then probably replace us in many of the activities we do. That is what robots and everything are being made for. So, that is, at a very basic term, what Artificial Intelligence is.
Machine Learning
Under Artificial Intelligence, we have Machine Learning. Machine Learning is a subset of AI which focuses on training a machine using algorithms. Machine Learning algorithms train the model to do certain tasks that are repetitive in nature. You feed and train the model, telling the machine what to do, and then that machine will slowly and gradually learn it and do it. For example, suppose you used to, on a daily basis, try to take data from multiple systems, multiple CRM systems, and then take it into an Excel sheet and then process it, understand and apply formulas and logic to take out marketing insights based on your understanding. If you put all that understanding into a model, give that model the logic and everything, then that whole activity which you used to do as a marketing analyst can be done by an AI service. I hope you get it.
Deep Learning
Deep Learning is a deeper subset of Machine Learning where we get into neural networks. What is a neural network? In our brain, everything is communicated through neurons. Deep Learning involves algorithms that use neural networks to understand human behavior. This is very important because it directly relates to ChatGPT. Using neural networks, the ChatGPT model—which is a subset, a sub-model, because there are various other models which I’m not covering—uses these neural networks because it helps a machine perform tasks such as image recognition. For example, if you give a machine with neural network algorithms, continuously showing it an image of an elephant, it will slowly and gradually start recognizing the features of the elephant and then clearly understand that it is an elephant and not a tiger.
What is ChatGPT?
ChatGPT is natural language processing. It allows the machine to understand what a human is speaking and then respond as if a human is responding. This involves text-to-speech or speech-to-text, among other things. I hope you get some gist of where we are now. In this world, a star has been born, and that star is ChatGPT.
Decoding ChatGPT
The Basics of ChatGPT
Let’s try to decode ChatGPT step by step. ChatGPT is a GPT model, which is a large language model used for building ChatGPT. GPT stands for Generative Pre-trained Transformer. Generative means it’s part of generative AI. Generative AI is a field where the model, if you use a generative model, will always generate human-like responses. It could be in terms of text or images. It generates something out of its own understanding.
Pre-trained
Pre-trained means that when OpenAI set up ChatGPT, they started feeding millions and billions of diverse internet data into this model. They pre-trained this model with lots of inputs and iterations, giving it different scenarios and teaching it how to respond to any query or question.
Transformer Model
Transformer is the actual model. If someone asks you which AI model is used in ChatGPT, you could say the Transformer model. ChatGPT is a large language model that uses neural network architecture. Neural network architecture, as discussed earlier, is similar to how brain neurons communicate. Input is sent through various hidden layers where complex calculations occur, and then an output is generated. All these hidden layers are not visible; you just give the input and get the output. For example, if you type a query in ChatGPT, you get a response, but behind the scenes, a lot is happening.
Other Products of OpenAI
The Transformer model is also used in DALL·E. DALL·E is a different product and platform given by OpenAI to generate images. For instance, you could type a description like “a dog with a cap on his head and sitting on a beach,” and it will automatically create that kind of image based on your input.
How the Transformer Model Works
The Transformer model was trained with a lot of diverse data to generate human responses. How does it analyze and respond like a human? The difference between a human and a robot is that humans work with their own weights and biases, their own thoughts and philosophy. These weights and biases need to be fed into the model to generate human-like responses. For example, if you ask me, “What’s your favorite pastime?” I could give you two or three different answers based on my mood. Similarly, the model uses a mechanism called attention management to decide the most important parts of the input and focus on them.
Attention management is like highlighting sentences in a book with a marker to give additional weight to a particular part of learning. The model acts like an orchestra or a symphony where a conductor instructs musicians to play specific notes to create beautiful music. The model’s neurons act in harmony to produce a coherent response.
Interaction with ChatGPT
When you interact with ChatGPT, it follows these five high-level steps:
- Input: When you write a question, it is sent as input to the ChatGPT model.
- Processing: The input is divided into words, then further subdivided into small units or chunks, and assigned tokens in a process called tokenization.
- Context Understanding: Using neural network architecture and attention management, the model understands the context of the input.
- Text Generation: New tokens are generated to prepare the output. Methods like sampling or beam search are used depending on the type of response needed.
- Output: The tokens are converted into readable text and shared with the user.
Real-Life Use Cases of ChatGPT
Now, let’s understand where ChatGPT can be used in real life. I’ll give some scenarios, but there are countless possibilities. Here are five use cases which have helped me personally and are relevant to people working in IT:
- Coding: People with coding backgrounds who don’t want to code can get snippets of code and modify them. Those with no coding background can input exact requirements and get Python or Java code generated, along with explanations of the logic.
- Technical Support: For technical troubleshooting, you can input problem descriptions into ChatGPT, which can provide insights and steps to follow.
- Job Search: ChatGPT can help in writing resumes and cover letters. You can feed job applications into ChatGPT, which can suggest what to add to your cover letter or resume.
- Content Creation: ChatGPT can assist in writing technical blogs, tweets, storytelling, YouTube video scripts, and more.
- Understanding Technical Concepts: ChatGPT can help understand technical concepts, providing specific responses that complement search results from Google.
These are just a few of the many ways ChatGPT can be utilized.
