Its multimodal neural network technology is currently in an invite-only testing phase.
The model can answer users’ questions, carry out a conversation, write a programme code to develop software, and create texts and images based on descriptions within a single context, the company said on Monday.
GigaChat supports “multimodal interaction and is better trained in the Russian language”, it added.
“Sber’s new solution that we have introduced today is breakthrough for the whole universe of Russian technologies,” said Herman Gref, president and chairman of the board at Sberbank.
“GigaChat is unique in its open architecture while the model’s global analogs stick to closed AI principle. It is important that GigaChat can be employed not only by mass users who like to experiment with innovation, but also by students and even scholars for research papers.”
Sberbank’s SberDevices — a manufacturer of smart devices founded in 2019 — and Sber AI teams participated in the model development with support from Moscow’s Artificial Intelligence Research Institute.
The global generative AI market is expected to reach $188.62 billion by 2032, growing at an annual rate of more than 36 per cent, from $8.65 billion last year, data from market research company Brainy Insights shows. The North American region has dominated the market this past year.
Launched in December, San Francisco-based research company OpenAI’s ChatGPT comes up with human-like responses to prompts in seconds, based on information publicly available on the internet. The programme has become hugely popular, while also raising concerns about what it is being used for and its accuracy.
Entering the race of generative AI, Google launched Bard in February.
In the same month, Microsoft announced its new AI-powered Bing search engine that aims to deliver better searches, more complete answers, a new chat experience and the ability to generate fresh content.
A number of experts have been engaged to ensure GigaChat has specific skills for the correct execution of instructions with full accuracy. The neural network has been trained on Christofari Neo, the second supercomputer of Sberbank launched in November 2021.
“GigaChat implements a number of solutions aimed at preventing the use of generative models that contradict the AI code of ethics … our team of developers will continue to refine the neural network’s capabilities with particular emphasis on ethics and security,” Sberbank said.
Sberbank’s virtual assistant Salut is one of the first products expected to shortly gain all the capabilities of the new model.
The lender has a distribution network of about 14,000 branches.