Organization laion (Large-scale Artificial Intelligence Open Network), developing tools, models and data collection for creating free machine learning systems (for example, collection Laion is used to teach the models of the STABLE Diffusion synthesis system), founded the project Open-SSISISTANT , in which the development of a chatbot with artificial intellect began The proprietary service of cheatgpt and capable of understanding questions in natural systems and dynamically extract the necessary information. The project code is written in Python and is distributed under the license Apache 2.0.
It is expected that Open-SSSISTANT will spur the development of open developments about the field of content formation and processing requests in natural languages, as the open project of Stable Diffusion stimulated the development of means for generating images. The project is aimed at creating an intellectual assistant to the future and does not plan to dwell on the repetition of ChatGPT capabilities. For example, in addition to such functions as writing messages and emails, it is planned to teach the bot to perform more significant work, use the API and dynamically investigate information, while providing funds for expansion and personalization. It is also planned to provide the most compact and effective solution that can be performed on ordinary consumer equipment.
The project develops a publicly accessible language model, trained in a large collection of data and enhanced through the application of training methods with reinforcement and attracting enthusiasts to obtain feedback on the quality of dialogs. At the first stage for training, it is planned to use at least 50 thousand examples of dialogs in the form of a request (instruction-films) prepared and reviewed with the involvement of the enthusiasts community.
In the future, to increase the efficiency of the system and avoid storing billions of predetermined parameters, they intend to use a dynamically updated knowledge base in the project that can extract the required information through search engines.
The project will also be provided with tools that allows you to use the prepared model without restrictions, to develop it taking into account its own preferences and create its applications and services on its basis.