ChatGPT : Magic Tech?

ChatGPT : Magic Tech?

Can a bot give u an intelligent story? Can one create code that works very well? Can one explain a concept? Give you an email? Well, what if I tell you that ChatGPT does all that and more, and that too for free(at least right now)!

Well, that's a lot of fuzz about ChatGPT, and a lot of people are asking what it is and how to use it. So let’s start with a bird's eye view of the topic, and expand it more.

To give you a very short overview, it is a chatbot capable of generating responses and content, based on user interactions, in a way that mimics an online chat, hence the name, ChatGPT.

In the 12 days since its launch (Nov 30th 2022), it has acquired over a million users and has been a worldwide sensation. Everyone is in awe of its stifling capabilities to produce convincing and fairly accurate text results for any query.

These capabilities did land it some hot water, with popular sites such as StackOverflow banning answers generated using the tool, due to relative inaccuracies.

ChatGPT generating publicity material for the article

Yes, the very copywriting you saw on social media was generated using ChatGPT, and a few fine tunings from my end. Let the AI give me some words to publicize an article about itself.😉😁

What is ChatGPT?

Chatbots have been with us for a long time, but this is something special. ChatGPT is the latest innovation developed by popular AI Research Company Open AI, and at the core, it is a chatbot based on GPT(Generative Pretrained Transformer) 3.5 language model.

The offering is initially in beta, and though it is still in the testing phase, its capabilities are fascinating. People are saying it can even replace GOOGLE, well that’s a bit of a long shot, isn’t it? Or is it? Well, what makes it special? What makes it stand out from the saturated chatbot market, where we see chatbots on every website?

To quote the official blog:

"We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests. ChatGPT is a sibling model to InstructGPT, which is trained to follow instructions in a prompt and provide a detailed response."

How Does it Work?

There are a ton of complex notes & guides that explains the nitty gritty of the nuances involved with the GPT3.5 flagship. To put it simply, the whole concept is a 3 step process

  1. Collect Demonstration Data and Train a Supervision Policy

    A prompt is sampled from the prompt Dataset, wherein a labeller demonstrates the desired output behaviour, which is then used to fine-tune GPT 3.5 with supervised learning.

  2. Collect Comparison Data and Train a Reward Model

    A prompt and several model outputs are sampled, and labellers rank the outputs based on their correctness. The data is then used to train the model.

  3. Optimize a policy against the reward model using the PPO reinforcement algorithm

    A new prompt is sampled from the dataset. The PPO(Proximal Policy Optimization) model is initialized from the supervised policy. The policy generates an output, and a reward score for the same is calculated using a reward model. The reward is used to update the PPO.

Official Graphic visualization from Open AI explaining the steps involved in ChatGPT training

Get a taste of the magic:

  1. Visit the website on your computer or mobile device

  2. Sign into your Open AI account, or sign Up if you are new.

  3. Enter a prompt, and viola, we can see the magic

  4. Not satisfied with the result? Well, we can always reset the conversation, by pressing Reset Thread

It’s amazing to see what an AI can conjure up currently, and it’s just the tip of the iceberg, a lot of amazing capabilities are going to follow.

Conversation with the bot

By the way, how does the bot answer questions about itself? Let's find out.

Let's start by asking if it knows itself, and what makes it unique.

Me: Explain ChatGPT? What makes it special?

ChatGPT: ChatGPT is a variant of the popular GPT-3 language model, which has been specifically fine-tuned for conversational understanding and response generation.

This makes it well-suited for tasks such as chatbots, dialogue systems, and other applications that require natural language understanding and response generation.

One of the key features that makes ChatGPT special is its ability to generate responses that are both fluent and contextually relevant to the conversation.

This is made possible by its large size and powerful training algorithm, which enable it to learn the nuances of language and generate responses that are appropriate for the given context.

Additionally, ChatGPT has been trained on a wide range of conversational data, which allows it to understand and respond to a wide variety of conversational styles and topics.

This makes it a versatile tool that can be used in a variety of applications, from customer service chatbots to social media bots.

The response was quite convincing and had a good level of self-awareness. Maybe we should ask whether it knows how it answers the queries ...

Me: How does ChatGPT answer the questions/queries? How does it generate these answers?

ChatGPT: ChatGPT uses a combination of machine-learning techniques to generate responses to questions and queries. First, it uses natural language processing (NLP) to understand the meaning and intent behind the input.

This involves analyzing the words and phrases in the input and using this information to identify the key elements and understand the overall meaning.

Next, ChatGPT uses its large body of pre-trained knowledge to generate a response that is relevant to the input.

This involves selecting the most appropriate information from its internal knowledge base and using this information to generate a response that is both fluent and contextually relevant.

Finally, ChatGPT uses a variety of techniques to polish and refine its generated response, such as grammar correction and rephrasing, to ensure that the output is natural and easy to understand.

Overall, ChatGPT's response generation process involves a combination of NLP, knowledge-based generation, and refinement techniques to generate high-quality responses to a wide range of input.

That is quite an immersive explanation and worded in the easiest way to understand revolutionary concepts. So there is still a valid question: Is it yet another chatbot, or is it different?

Me: How do you differ from a chatbot?

ChatCPT: I am a virtual assistant trained using natural language processing (NLP) and machine learning algorithms, whereas a chatbot is a type of software that is specifically designed to simulate conversation with human users.

I am able to provide more accurate and sophisticated responses because I have been trained on a large body of text data and use advanced machine learning algorithms to generate my responses.

Additionally, I am not limited to a specific set of pre-programmed responses like most chatbots. Instead, I am able to understand and respond to a wide range of input, using my pre-trained knowledge and NLP capabilities to generate appropriate responses.

This allows me to provide more useful and relevant information to users, and to engage in more natural and fluid conversations.

Some more examples

Let’s have a look at some of the examples, such as email generation, text summarisation, article creation, poetry creation, presentation guidelines, etc.

  1. Let's start by asking ChatGPT to generate a work email, requesting work from home, and a change in work hours to your manager.

    ChatGPT generated work email, asking for Work from Home

  2. How would the AI explain Radioactivity to a curious 10-year-old?

    ChatGPT explains Radioactivity to a 10 year old

  3. Maybe it can help to generate a college-level presentation on the same topic.

    ChatGPT's guidelines on how to create a college level presentation on Radioactivity

  4. Let's end with some poetry about goals and life

    A poem written by ChatGPT on Life and Goals

Well, the results seem promising, doesn't it? Maybe we are heading to the start of AI-assisted content generation. Let's wait and watch!