General models such as chatGPT have two limitations: First, they do not contain any additional information after the cut off date, and second, they cannot answer questions about specialized fields other than the general information utilized for training.
Therefore, RAG builds a separate DB related to this specific expertise. To answer a user’s question, LLM engine utilizes it first, and additional knowledge is supplemented by its previously learned knowledge or internet searches. This has become acknowledged as advantageous way to respond to specialized knowledge in a specific field.
TONchat is a Q&A service based on the chatGPT API. Basically, It stores various kind of information in a database and utilizes them as response materials : Smart contracts developed by Tokamak Network and their related annotations, as well as various utilization documents formed by the community, such as user guides and Q&A documents.
This configuration is commonly referred to as a RAG (Retrieval Augmented Generation) application. As LLM (Large Language Model) based AI is rapidly growing, RAG is one of the application services that utilizes LLM to overcome its limitations.
As soon as its completion, we plan to transfer all rights to the development and use of TONchat (including commercial utilization) to the Tokamak Network community. This will allow various knowledge and know-how generated within the community to be accumulated and easily shared, not only for the convenience of existing users but also to remove obstacles to the influx of new users.
TONchat has only one goal. To make the source code of Tokamak Network more accessible to the community.
TONchat will open the alpha and beta versions in the first quarter of 2024, followed by updates in the CI/CD field, such as the automatic collection of learning data, and user feedback, and will complete development in the fourth quarter of 2024.
