Generative Pre-trained Transformer 3 or hereinafter referred to as GPT3 is a model from the field of Machine Learning developed by OpenAI. The way GPT3 works is predicting or generating word for word to complete a series of sentences from the Prompt instructed by the User.
GPT3 has now undergone several upgrades/enhancements, where now the default model is already using version GPT3.5 or commonly known as InstructGPT. This model is better at recognizing "instructions" or commands from the User than the previous model.
In order to be able to generate words and understand their context in natural human language, GPT3 is trained using a data set in the form of a very large text. The estimated size of the data used is ~750GB which is adapted from Common Crawl, Wikipedia, various collections of books, journals, and other Internet references.
The way it works so that we can use this prompting feature can only be accessed via the API endpoint that has been provided by the developer (OpenAI). Because GPT3 is no longer open-source and imposes pricing on the use of the model.
GPT 3 consists of 4 model choices, where for the Chatbot case the model used is the highest class named Davinci.
By combining Botika current design of the Chatbot system, Rule-based Technology, NLP Intent Classification with Generative Pre-trained Transformer 3 (GPT3) So that it will produce dialogue through chatbots that are more natural, accurate and fast response for business conversations.
Luna sudah menggunakan Teknologi GPT-3 dari Microsoft OpenAi Service