Russia’s Sberbank Launches AI To Compete With ChatGPT

The Russian state-owned company Sberbank on Monday launched its own artificial intelligence (AI) chatbot GigaChat as its answer to OpenAI’s ChatGPT.

Taking a shot at ChatGPT, Sberbank said, “Unlike the foreign-made neural network, GigaChat was initially designed to support multimodal interaction and communicate in Russian more intelligently.”

The company said GigaChat is going to be available in a pilot mode and can be joined by invitation only. It can answer questions, maintain dialogue, write program code, create texts and pictures based on descriptions within a single context.

The GigaChat architecture is based on the neural network ensemble model called NeONKA (Neural Omnimodal Network with Knowledge-Awareness), which includes several neural network models along with supervised fine-tuning and reinforcement learning with human feedback.

Sber’s new neural network is capable of addressing many intellectual tasks: supporting a conversation, writing texts, and answering fact-based questions.

By incorporating the Kandinsky 2.1 model into the ensemble, the neural network can also create images. “We plan to make NeONKA 3.5 with 13 billion parameters publicly available to accelerate the evolution of the open-source community and artificial intelligence. It is of great importance for us to enable as many talented students, scientists and developers as possible to use the latest available technology,” Sber said.

The company said that the model is being widely introduced into many Sber products and services, which will become even more handy for its clients. ‘Salute’ the virtual assistant will be among the first products to take advantage of everything the new model has to offer.

The first version of NeONKA has been optimized in terms of size and speed to fit high load systems.

Trained using the Christofari Neo supercomputer, the model is being developed by SberDevices and Sber AI with support from AIRI — Artificial Intelligence Research Institute and some AI experts recruited to train GigaChat in some specific knowledge domains, fine-tune it to perform correctly and accurately.

In a statement, Herman Gref, CEO and chairman of the Executive Board, Sberbank said, “The all-new Sber design we have presented today is a breakthrough for the entire and large universe of Russian technology. Moreover, GigaChat is unique in that it has open-source architecture, whereas its global equivalents pursue a Closed AI strategy. Importantly, GigaChat can also be used by students and serious researchers in their papers, as well as by mass users who like to experiment with innovation.”