Ideas2IT rewards key players with 1/3rd of the Company in New Initiative.  Read More >
Back to Blogs

ChatGPT - Closer now to J.A.R.V.I.S

Just like Tony Stark talking to J.A.R.V.I.S in Iron Man movies, Thanks to OpenAI we can also start talking with computers which could understand the context and respond to us better than simple chatbots. It is claimed that it can respond to everything - from stories, mathematical solutions to essays and also can create poems and content on various topics too.

What is ChatGPT and How it came to be:

After GPT3 which is a transformer-based NLG model, OpenAI was focusing towards conversational solutions. They already had prior work done in their InstructGPT but InstructGPT is just a prompt-based QA model, it lacked the ability to perform a full-blown conversation.Perfecting what their prior works lacked, OpenAI presents ChatGPT, which can perform fully context-aware conversations making it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.

How is ChatGPT different from other Chatbots?

One of the main differences between ChatGPT and a typical chatbot is the way they generate responses. ChatGPT uses the latest version of OpenAI's GPT-3 language model, which is a powerful and sophisticated machine learning model that has been trained on a vast amount of text data. This allows ChatGPT to generate responses that are more human-like and natural, as it has been trained to understand and generate human-like language.On the other hand, a typical chatbot is often limited to pre-programmed responses based on a set of rules or keywords. This means that it can only respond to specific queries or commands that it has been trained to recognize, and it may not be able to generate responses that are as natural or human-like as those generated by ChatGPT.Another key difference is the level of flexibility and adaptability. Because ChatGPT is powered by GPT-3, it is able to understand and respond to a wide range of inputs and is able to adapt to different conversational contexts. This allows it to handle more complex and nuanced conversations and provide more useful and relevant responses. In contrast, a typical chatbot is often limited to a specific domain or set of topics, and may not be able to handle unexpected or out-of-domain inputs as effectively.Overall, the use of GPT-3 makes ChatGPT a more powerful and sophisticated chatbot compared to a typical chatbot, allowing it to generate more human-like responses and handle a wider range of inputs and conversational contexts.

Other similar works in the same area?

This may not be the only bot that offers such services. Several such models exist, for instance, we had Microsoft’s AI chatterbot named ‘Tay’ - ‘Thinking About You’ which was released on Twitter in March 2016, official name TayTweets, and its handle was @TayandYou. But it was soon taken down after it posted racist, sexist, and offensive comments which it had learned from Twitter users.Chinchilla from deep mind - 70 billion parametersMegatron-Turing NLG from Microsoft - 530 billion parametersBlender bot 3 from Meta (175 billion parameters) reportedly replied to some bad things about the CEO as a response to a prompt stating that his company exploits people for money and he doesn’t care. It needs to stop!Thus ChatGPT’s USP is that it can filter out inappropriate content and choose to provide an advisory answer for all such content, Thanks to OpenAI careful thoughts and time spent during the training phase.

How does ChatGPT work:

How does ChatGPT work:

It uses a RLHF process. RLHF stands for Reinforcement Learning from Human Feedback.As you can see from the above diagram the steps 1 to 4 were done for many iterations on various subjects and topics for a very long period of time which made this model perfect.Now the steps 1 through 4 are being repeated as we speak to make the model much better.The training based on steps 2 and 3 is what is called RLHF.

How to use this system:

Potential users and people who would like to research the application can try to sign up to ChatGPT and use it for free upto 900,000 tokens.At the current phase after signing up you can try and use the transformer and the conversations we make will be used by the OpenAI team to further train their transformer to be much better in the long run.Some examples of questions you could ask maybe:

  1. Explain quantum computing to a teenager.
  2. Give me an idea for my Kid’s birthday.
  3. How do I make an HTTPS connection in flask.

Sample Q and A:Is the world perfectly spherical ?“No, the world is not perfectly spherical. It is an oblate spheroid, which means it is slightly flattened at the poles and bulges at the equator. This is due to the Earth's rotation, which causes the planet to bulge at the equator and flatten at the poles.”

Can ChatGPT be Fine-Tuned?

Soon but not immediately, We can fine-tune ChatGPT to improve its performance on specific tasks or domains. Fine-tuning is a common technique in machine learning where a pre-trained model is adapted to a new dataset or task by updating its parameters using additional training data. This allows the model to better capture the patterns and characteristics of the new data, and to improve its performance on the specific task or domain.Fine-tuning ChatGPT would involve providing it with additional training data specific to the task or domain you want it to perform well on, and then running the fine-tuning process to update its parameters. This can be done using the tools and libraries provided by OpenAI, such as the GPT-3 API. Fine-tuning ChatGPT can help improve its performance and make it more effective at handling specific tasks or domains.


Context Awareness:

Context Awareness - chatgpt

Here in the second conversation I just told it and it found that I am talking about the world.

Emphasis on Safety and Ethics:

Emphasis on Safety and Ethics:


Limitations -chatgpt

The one limitation we can observe is it may not know of topics like movies and in movie context (Interstellar released in 2014), and other information it provides is it is trained till events of 2021 hence limited to that data.

Ideas2IT Team

Connect with Us

We'd love to brainstorm your priority tech initiatives and contribute to the best outcomes.