Yandex has developed Sona, a generative AI model designed to streamline recommendation systems by consolidating candidate generation and ranking into a single transformer model. This approach eliminates the need for multiple, separately trained models typically found in recommendation cascades. Sona was tested on Yandex Music smart speakers, where it replaced over 15 candidate generators and ranking stages. The results showed a significant increase in user engagement, with a 4.53% rise in active users and an 11.42% increase in likes during a live production experiment. AI
IMPACT This approach could streamline recommendation systems across various platforms, potentially improving user engagement and content discovery.
RANK_REASON New generative AI model release from a major tech company (Yandex) with performance metrics.
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →