The next big thing ChatGPT

ChatGPT is a chatbot developed by OpenAI and released in November 2022, it comes from the GPT family of OpenAI. Here GPT stands for Generative Pre-trained Transformer, as suggested from the name it is a pre trained bot and comes with learning techniques. As AI is the next future of our world a lot of focus is put on AI and developing it.

In simple terms you can say ChatGPT is a google alternative but compared to google where when we search anything it gives us a list of websites whereas in ChatGPT we get a detailed response and very articulated answers across various domains. Though it seems to be very interesting as it is considered to be the next big thing it is just a prototype. It has certain drawbacks as there were uneven discrepancy in the responses.

In order to train the AI, human trainers were used in both sides such as users and an AI assistant. The conversations were then provided to the model which then improved its supervised learning. In another learning called reinforcement learning, the responses from the previous conversations were ranked. The ranked responses were then fed to the AI as reward models which were then fine tuned with the help of PPO which is called Proximal Policy Optimization. Apart from the above approaches OpenAI gathers data from the users of ChatGPT so that the AI can be fine tuned and trained. Users are asked for feedback on the improvements required, drawbacks present and also provide information on the best features.

Certain features of the ChatGPT are

· It can compose music on its own

· It can write and debug computer programs

· Write essays, fairy tales

· Answer test questions

· Behave similarly as a Linux system

ChatGPT tries to remove harmful and incorrect responses, the AI tries to gather information of the topic which has been asked and then as per the correct results it provides a detailed analysis of the search. There is also an API present which restricts ChatGPT to provide results on racial prompts or sexual prompts.

In December 2022, it is said that more than 1 million people use ChatGPT and are very happy with the results but with every invention there are drawbacks as well. It is said that as there is human intervention in the training so there can be over optimization of the AI and thus can cause disruptions in the performance. It has been mentioned that any answers from ChatGPT can be trusted or not. Some even found that ChatGPT can be used to create malwares and write phishing emails which can hamper security. There are people who were able to bypass the content policy present in ChatGPT which is harmful to the chatbot itself.

From the earlier times it has been seen that with every invention there were negative reviews so it does not change for ChatGPT as well. A lot of journalists have tested it and provided the negative feedback. OpenAI is working on it as it considers it to be a path breaking which can revolutionise the way we currently use internet. So, there is more focus in fine tuning ChatGPT and hope it will become worthy of its hype.