Russian AI breakthrough speeds up scientific research with 2.7x efficiency

Researchers at HSE University have improved how large language models handle Russian scientific terms. Their new approach makes AI faster, more accurate, and far more efficient in specialised fields. The changes could help automate complex scientific analysis in the future. The team fine-tuned existing AI models using the iFORA-QA dataset, which contains carefully checked reports on science, technology, and innovation. After adaptation, the system answered highly technical questions with greater precision.

Speed and efficiency saw major gains. The updated AI now responds 2.7 times faster than the original open-source model. Memory demands also fell sharply, requiring 73% less capacity to run. Further improvements are planned. The researchers intend to add tools that reduce the risk of 'hallucinations'—instances where AI generates incorrect or misleading information. Their long-term goal is a full system of intelligent agents designed specifically for Russian scientific needs.

The adapted AI now performs better in specialised fields while using fewer resources. Faster response times and lower memory use could make it a practical tool for researchers. The next phase will focus on refining reliability and expanding its role in scientific workflows.