ChatGPT, AI wonder machine that’s taken internet by a storm: Here’s how it works

As per the company, the AI platform has been trained using the Reinforcement Learning from Human Feedback (RLHF). In simple words, the human AI trainers initially provided conversations in which they played both the AI assistant and the user. Moreover, the company had given access to trainers about model-written suggestions which helped the trainers to compose their responses.

Moreover, OpenAI created a reward model for reinforcement learning. This means that the company wanted to collect some comparison data for better answering mechanisms. To collect such data, the company utilised the conversations between AT trainers and chatbot. The company selected a mode written message, sampled several alternative completions and the AT trainers ranked these messages. Hence, it is trained to follow an instruction in a prompt and provide a detailed response. Users can simply feed in their query and the chatbot will reply to them. But then how is it different from other Artificial Intelligence (AI) chatbots? As per the creators, ChatGPT, unlike other AI chatbots, can answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.

As per OpenAI, ChatGPT is a free service, but only during the research preview. This means one can simply head to OpenAI website and click on the Try ChatGPT button to begin using the platform. You can either sign up or use your OpenAI account to start using the ChatGPT. The company has also given a sample on the web page for reference.

ChatGPT is the latest technology in the Generative Pre-Trained Transformer (GPT) family. To put in simple words, it is the latest tool in auto text-generating AIs. But, it is not free from errors or limitations. OpenAI, on its website, admits that ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers. Also, the model often excessively verbose and overuses certain phrases.