Yandex has announced a notable achievement in the realm of language models: YandexGPT 2 has outperformed OpenAI’s GPT 3.5 in understanding and responding to queries in Russian.
According to a recent study conducted by Yandex specialists, YandexGPT 2 delivered superior answers in 63% of cases when compared to its OpenAI counterpart.
The methodology employed in this study was the Side by Side (SBS) comparison method. Both YandexGPT 2 and GPT 3.5 were tasked with responding to identical queries, and AI trainers, unaware of which model generated which answer, chose the best response.
This approach ensured an unbiased evaluation of the models’ performance. The queries used in the study were sourced from anonymous user requests to YandexGPT 2, encompassing a wide range of topics and reflecting the genuine needs of people using Yandex services.
Alexey Gusakov, the technical director of Yandex Search, emphasized the practical importance of such comparisons.
He stated, “Comparisons of neural networks are needed not to obtain abstract numbers in a benchmark, but to solve product problems, create new functions and develop business. That is why a significant part of the comparison basket was made up of real requests from our services.”
This achievement follows the launch of the improved YandexGPT 2 model in September.
The updated model boasts enhanced capabilities in handling a broader spectrum of problems and has shown a significant improvement in the quality of its answers.
In early December, Yandex introduced a new version of its voice assistant powered by YandexGPT 2 and opened applications for its beta testing.
This development represents a significant stride in Yandex’s commitment to advancing AI technology and enhancing user experience.
Yandex’s success with YandexGPT 2 in the Russian language domain marks a pivotal moment in the AI landscape, demonstrating the potential for region-specific models to outperform global counterparts in localized contexts.
Source: Yandex Press Service