In a recent comparative assessment, it was reported that the Russian neural network YandexGPT 2 provides higher quality responses in Russian than OpenAI’s GPT-3.5 in 63% of cases. The information came from the Yandex press service and was shared with socialbites.ca.
The evaluation used a side-by-side method. Both neural networks received identical questions, and AI trainers selected the best response. Importantly, the evaluators did not know which system produced each answer during the process.
The question pool originated from anonymous user inquiries submitted to Yandex GPT 2. A diverse sample was created to reflect real user needs across a wide range of topics, ensuring broad coverage and applicability of the results.
Alexey Gusakov, the Technical Director of Yandex Search, explained that benchmarks serve a purpose beyond mere numbers. He emphasized that benchmarks should help address product challenges and guide the development of new features rather than exist as isolated metrics.
YandexGPT is a neural network developed within the Yandex GPT family. It generates text by drawing on data from the Internet and already includes built-in capabilities compatible with the Alice voice assistant. The project underscores Yandex’s ongoing work to integrate conversational AI into practical products and services.
The comparison and ongoing development signals a broader strategy to advance AI-driven text generation for Russian-speaking users and to explore how such tools can augment everyday digital tasks and information retrieval. [Citation: Yandex press materials]