AI is the milk and bread of today’s world. We can find AI being used in everything we use in our day-to-day life. Google, Amazon, IBM, Microsoft, and all the Silicon Valley companies invest million and billion of dollars to perfect AI and transform our lives in a way that we can’t imagine. One such AI tool is ChatGPT developed by OpenAI which is taking people by storm, gaining millions of appreciation within a few months of its launch.
Introduction of ChatGPT
Firstly, let’s see what ChatGPT has to say about itself.
ChatGPT also known as Chat Generative Pre-Trained Transformer developed by OpenAI based on the GPT-2 and GPT-3 family of large language models. Sam Altman, the president of OpenAI, is the creator of ChatGPT which is fine-tuned by both supervised and reinforcement learning techniques. It was launched on 30th November 2022. OpenAI aims at creating and promoting friendly AI responsibly.
ChatGPT is trained to be conversationalist with the user at the other end. It understands the question’s reference, and conceptuality and then gives human-like answers. It works on an interactive model which can keep track of previous questions asked and also understand the reference given based on previous questions.
Latest Trends
Interest in OpenAI’s ChatGPT skyrocketed in recent weeks.
In the below graph, we can see the rate at which ChatGPT is trending in the entire world.
And here is the google search result from 7th January 2023
and 10th February 2023.
We can see the difference and guess the rate at which people are writing and discussing on ChatGPT.
Underlying Technology and Training
ChatGPT is based on GPT-3 which is a state-of-the-art family of the large language model. It is the most powerful language model with 175 billion parameters. It is fine-tuned on top of GPT-3 using supervised learning and reinforcement learning techniques. Both techniques require the human trainee to improve performance.
In supervised learning, a human performs the role of both the AI and the user. In reinforcement learning, the responses from previous conversations are ranked. Using these ranked responses a reward model is created which is further fine-tuned by using Proximal Policy Optimization (PPO). It is a cost-effective algorithm model trained in collaboration with Microsoft on their Azure platform. It negates many computationally expensive operations with faster performance.
GPT-3 model was trained by providing data from textbooks, web text, articles, Wikipedia, and vast vocabulary which round up to 570GB of data or to be more exact 300 Billion words are fed to it. Using all the training and knowledge base ChatGPT can provide human-like answers, have a friendly conversationalist structure, and understand concepts of conversations.
How to Use ChatGPT?
Using OpenAI ChatGPT is very easy. You just have to register yourself in OpenAI and then start asking your long list of questions. Open Ai has not yet released any mobile or desktop app, so it cannot be downloaded. You have to visit the site to use it.
- Go to the Open AI site. You will land on the login and registration page.
- You have to first register yourself by providing the email or you can also log in using your google or Microsoft account.
- After giving an email next provide a password and a verification link will be sent to your email id.
- After verifying your email id, you have to give your name and mobile number.
- You have to verify your mobile number by entering a code sent to your number.
- Now you are ready to use ChatGPT.
Features of ChatGPT
At the basic level, ChatGPT is a chatbot but a versatile chatbot. It is limited to text only same as its brother DALL-E can produce images only. Let’s dive into its features:
- Imitate real human-like answers. Using its deep learning and natural processing language to provide natural answers by maintaining the integrity and relevance to the question.
- It can provide answers to complex problems by understanding them correctly within no time whereas humans may spend days understanding the question and generating answers.
- It can help you write academic articles, literary texts, poems, jokes, song lyrics, or even continue your poem.
- With the help of training data related to programming languages like python and bulletin board systems, it can write codes or debug the errors in the written code.
- It is highly customizable through which people can customize their chatbot as per their needs.
- It can reply in various languages.
- It can stimulate an entire chat room, or an atm, emulate a Linux system, and even play games like tic-tac-toe.
- Unlike its predecessor InstructGPT, ChatGPT attempts to reduce harmful and deceitful responses.
Where ChatGPT fails?
ChatGPT is an impressive and advanced text-based AI tool but it also has some drawbacks which can’t be ignored. In OpenAI’s own words:
“ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers. Fixing this issue is challenging, as: (1) during RL training, there’s currently no source of truth; (2) training the model to be more cautious causes it to decline questions that it can answer correctly; and (3) supervised training misleads the model because the ideal answer depends on what the model knows, rather than what the human demonstrator knows.”
Another that’s important to highlight:
“While we’ve made efforts to make the model refuse inappropriate requests, it will sometimes respond to harmful instructions or exhibit biased behavior. We’re using the Moderation API to warn or block certain types of unsafe content, but we expect it to have some false negatives and positives for now. We’re eager to collect user feedback to aid our ongoing work to improve this system.”
Apart from the above-mentioned limitations of OpenAI themselves, now let us draw out the limitations that we found.
- The software has limited knowledge of any world activity that happened after 2021.
- It is impossible to keep up with all the world activity as it happens and provide training data to AI.
- Both training data and bias in data can create a negative output model as it has shown bias when it comes to training minority data groups.
- The chatbot can generate incorrect answers when questions asked include too many factors and are really niche. The system becomes overwhelmed and completely ignores certain prompts.
Conclusion
After a decade the tech world has seen the advent of new AI which has blown the public mind and has become a phenomenal success in such a short time. Within just five days ChatGPT has reached a 1 Million user base and in two months it has crossed 100 million users.
There is no doubt it will revolutionize the tech industry by lips and bounds. Some major companies have already started to build their own similar AI as an answer to the challenges that ChatGPT has brought into the market. It is just the beginning of something new and big.
Thank You for sticking with us till the end. Some other title you may have interest in is Benefits of AI.