How ChatGPT Works

How ChatGPT works, and what the steps are that its creators had to go through.


“Artificial Intelligence – Resembling Human Brain” by deepakiqlect is licensed under CC BY-SA 2.0.

By: Salar Nasimov, Journalist

ChatGPT is an AI chatbot that can answer questions and assist you with tasks.  But many people have been wondering how it works.  There are 4 phases, Pre training, Fine Tuning, Architecture, and Response generation. To begin with, pre-training is when the AI is trained with a large amount of text available on the Internet. It learns to predict future words based on what is has already wrote. This helps the AI understand language and learn different connections between words. After pre-training is fine tuning. The AI learns how to specialize for a task or data, it does this by using a database of chat data that is previously provided. The model learns from this data to generate meaningful responses to given questions. Next is architecture. ChatGPT is based on transformer architecture. Transformers are learning models that can capture sequential data such as text. They consist of an encoder-decoder structure with many layers of self-checking that allow focus on different parts of the inputted text when generating responses. Finally is Response generation: When someone enters a message, ChatGPT processes the input and makes a response based the paters and information he learned from the training data. An answer given and each word depends on the previous one.  So if one word changes, the rest after will change as well.


RELATED STORIES:,and%20produce%20high%2Dquality%20responses.