MIT and Google Launch StableRep Technology to Train Efficient AI Models Using AI-Generated Images


AntGroup introduced a multilingual multimodal large model training framework at the Hong Kong FinTech Festival, breaking through the bottlenecks of multilingual applications. This technology targets small languages such as Egyptian Arabic, and through a language-aware optimization framework, it achieves a 'thinking in the target language' mechanism, improving the training effectiveness for resource-scarce languages.
Apple faced obstacles in developing its own Siri large model, so it turned to a collaboration with Google, adopting a customized Gemini language model to enhance AI capabilities. The new strategy will adopt an 'edge-cloud collaboration' hybrid model, combining the advantages of cloud-based large models with local data privacy protection, aiming to optimize user experience and address shortcomings in handling complex tasks.
Microsoft Azure's ND GB300v6 VM set a new record of 1.1M tokens/sec for Llama2 70B inference, powered by NVIDIA's GB300NVL72 system with 72 Blackwell Ultra GPUs and 36 Grace CPUs, showcasing Microsoft's AI scaling expertise.....
Modern Motor and NVIDIA deepen their cooperation to build an AI factory based on the Blackwell architecture. The two companies announced joint development of projects in autonomous driving, smart factories, and robotics at CES. The project has received support from the South Korean government and will be detailed at the 2025 APEC Summit in South Korea.
The 2025 Hong Kong Fintech Week focuses on the integration of fintech and AI, bringing together guests such as Carrie Lam and Geoffrey Hinton. Zhu Guang, CEO of Du Xiaoman, emphasized the innovative applications of large models in financial services, driving customer service from monthly surveys to real-time responses, achieving a revolutionary transformation centered around customer-centricity.