The Ministry of Digital Development and businesses are discussing grants and subsidies for creating Russian-language and industry-specific datasets to train large AI models. A decision on support measures is planned to be agreed upon by September 30.
The idea is that one high-quality dataset could be used by several research teams. This should reduce redundant data collection, cleaning, and labeling.
However, experts warn of another problem. Without requirements for further use, large companies could receive state funds, and datasets created with budget money would remain closed.
According to Zaur Abutalimov, CEO of Ivideon, a typical custom video analytics project for the company takes 4–5 months, and up to 80% of this time can be spent on data collection, cleaning, labeling, and verification.
The cost depends on the task and can range from 1 million to tens of millions of rubles. However, the parameters of future grants have not yet been determined: the budget, participant requirements, and the share of state funding are unknown.
The idea of funding datasets is conceptually correct, but the money might go to 'giants' without a requirement to publish the results.
The total planned state support for AI is estimated at approximately 71.7 billion rubles. However, datasets cover only part of the needs of AI projects: they also require computing power, specialists, and infrastructure.