#ChatGPT- What Is It and How Does It Work Exactly?

Social networks gushing about it, school and college kids burbling about it, and professionals praising it: what’s so special about this ChatGPT? Read on to find out.

#Bots, #AI tools, @chatbots, and #virtualassistants have been making our lives easier for over a decade now. Be it customer services, sales inquiries, or troubleshooting, chatbots have become an integral part of life in today’s world. #ChatGPT is the latest entry, making waves in the world of #ArtificialIntelligence. This is a very versatile chatbot launched by OpenAI in November 2022.

But, what does this chatbot do differently that’s made it go viral in such a short time? Let us see what ChatGPT really is, how it works, why it has become the “next big thing” in the world of technology, and what we should be aware of while using it.

What is ChatGPT?

ChatGPT or Chat Generative Pre-trained Transformer is a Natural Language Processing (NLP) model built on the GPT-3 family of large language models. It has the ability to conduct conversational dialogues that appear human. It is a Large Language Model (LLM), trained with huge volumes of data to accurately predict what word comes next in a sentence.

In layman’s terms, ChatGPT has an in-depth understanding of language, be it spoken or written, and has many applications in the real world. It can summarize an essay, write research papers, translate content, and even write poems. It can explain many complicated topics in plain or technical language, depending on our preference. It can even write code and debug it. And all of this in a matter of seconds.

How Does ChatGPT Work?

ChatGPT was trained using Reinforcement Learning with Human Feedback (RLHF). The chatbot basically works like an autocomplete function but at a much grander scale. It uses a multi-layer transformer network, an effective deep-learning architecture for processing natural language to generate relevant responses.

Here’s how it works:

• Baseline Model

The initial model was trained using supervised fine-tuning. Human AI trainers provided conversations in which they played both the user and the AI. These trainers based their responses on model-written suggestions to make them more authentic. This dataset, along with a transformed dataset from InstructGPT (another chatbot by OpenAI) comprised the data for the training of the baseline model.

• Reward Model

Reinforcement Learning (RL) uses a reward model that helps the initial model improve. The trainers were given a prompt and several sample alternative responses to it and were asked to rank them from best to worst. This ranking data was used to train the reward model.

• Proximal Policy Optimization (PPO) Model

PPO is an advanced RL algorithm that constantly learns from and updates the current policy, (the strategy the model uses to achieve its goals) rather than past experiences. It ensures that no large changes are made to the policy and that the training is more stable. A new prompt is sampled from the dataset and the PPO model is now initialized from the supervised policy. This policy now generates the output. The reward model calculates a reward for this output and this reward is used to update the policy using PPO.

These are the three broad steps involved in the creation and working of ChatGPT. It takes in every input, processes them using its neural network architecture, and produces a response that is relevant to the context.

Why the Hype?

Probably the biggest cause for ChatGPT’s popularity is how it was made available to the public in a way that they could understand. Its conversational outputs, ability to answer follow-up questions, challenge incorrect arguments, admit mistakes, identify sentiments, and ability to simplify even the most complex of subjects are what make it stand out. It gained over 1 million registered users in five days and became the fastest-growing tech platform ever.

What to Watch Out For?

?One thing you don’t have to watch out for any time soon is this AI taking over the world and making humans its slaves. At least not anytime soon. Jokes aside, some real concerns that you will have to keep in mind while using ChatGPT are:

• It can sometimes give responses that sound plausible but are factually incorrect or nonsensical.
• It is sensitive to input tweaks. Rephrasing a query could get you a different answer or even no answer.
• It can be excessively wordy and repeat certain phrases.
• It misinterprets ambiguous queries instead of clarifying them.

To Conclude

ChatGPT is like no other piece of technology. It is quick, interactive, intuitive, and human-like. But it is also like every other technology in that, it has its limitations and there is always the risk of misuse. So, it is up to each individual to judiciously decide how, when, and where to use it. While it’s great to use state-of-the-art technology in your daily life, don’t forget to draw the line somewhere. After all, it is up to us to decide whether technology is going to be a useful servant or a dangerous master.