Sou Hu Cai Jing·2025-08-31 18:36Core Insights - Microsoft AI has announced significant advancements in artificial intelligence by launching two self-developed AI models, marking an important milestone in its AI technology journey [1][6] Model Details - The first model, MAI-1-preview, is an end-to-end trained foundational model, while the second model, MAI-Voice-1, is a voice generation model that offers high-fidelity audio and extensive customization options [1][3] - MAI-Voice-1 allows users to select emotional modes, voice templates, and up to 40 different speech styles, providing a rich auditory experience [1][3] - MAI-1-preview is a mixed expert model that has been pre-trained and fine-tuned on approximately 15,000 NVIDIA H100 GPUs, focusing on instruction following and everyday question answering capabilities [3][4] Deployment and User Experience - MAI-Voice-1 is already available on platforms like Copilot Daily and Podcasts, although it currently does not support Chinese output [3] - MAI-1-preview will be integrated into some text scenarios within Copilot in the coming weeks to gather feedback and enhance user experience [3][4] Strategic Direction - Microsoft AI CEO Mustafa Suleyman emphasized the importance of self-developed models for ensuring choice and maintaining a proactive stance in future developments, while continuing collaborations with companies like OpenAI [3][4] - Suleyman highlighted the focus on efficiency and high-quality training data in model development, aiming to maximize the utility of each computation [4] - Microsoft AI has a five-year roadmap with quarterly investments, anticipating the emergence of millions of AI models with diverse personality traits across various fields [4][6]