Kakao Unveils Advanced Multimodal Language Model and Specialized MoE Language Technology

Kakao has launched two advanced open-source AI models: the multimodal model “Canana-1.5-v-3b” and the specialized MoE model “Kana-1.5-15.7b-a3b.” These developments align with the Korean government’s push for Sovereign AI and emphasize Kakao’s commitment to independent technology. The Canana model excels in understanding both text and images, outperforming global counterparts in Korean benchmarks and showing competitive capabilities in English. The MoE model activates only a fraction of its parameters during inference, optimizing performance and resource use. Kakao aims to enhance AI’s human-like reasoning through ongoing development and open-source accessibility, which boosts competitiveness in the AI ecosystem. Despite facing delays compared to rivals like ChatGPT, Kakao’s robust data and service planning offer significant potential. However, the initial public reception has waned, with challenges in integrating AI functions into Kakao Talk. Users are eager for a more cohesive interface and comprehensive service expansion in the near future.

Source link