**联发科推出创新AI平台“达哥”与MR BreeXe，繁体中文能力赶超GPT-3.5**
在今日的生成式AI论坛上，全球知名芯片制造商联发科（MediaTek）宣布了其新一代生成式AI服务平台——MediaTek DaVinci，昵称为“联发科技达哥”。该平台的发布标志着联发科在AI领域的又一重大突破，目前已吸引超过40家厂商加入其生态系统，共同探索AI的无限可能。
在此次活动中，联发科还推出了其最新的繁体中文大型语言模型——MediaTek Research BreeXe。该模型基于Mixtral 8x7B模型构建，具备节省运算资源和提升处理速度的显著优势。据官方宣称，MediaTek Research BreeXe在繁体中文测试项目TMMLU+和MT Bench TW中表现出超越GPT-3.5的卓越性能，这无疑为AI语言处理领域树立了新的标杆。
MediaTek DaVinci平台和MR BreeXe模型的推出，不仅展示了联发科在AI技术研发上的深厚实力，也预示着未来AI在繁体中文应用场景中的智能化水平将进一步提升。联发科的这一系列创新举措，将有助于推动全球AI产业的发展，尤其是在中文语言处理领域的进步，为用户提供更加智能、高效的交互体验。
英语如下：
**News Title:** “MediaTek Stuns with Launch: DaVinci and MR BreeXe, a New Benchmark in Traditional Chinese AI超越GPT-3.5”
**Keywords:** MediaTek DaVinci, MR BreeXe, Surpasses GPT-3.5
**News Content:**
In today’s Generative AI Forum, renowned chipmaker MediaTek announced its groundbreaking AI service platform, MediaTek DaVinci, affectionately nicknamed “MediaTek DaGe.” This launch signifies another major milestone for the company in the AI domain and has already attracted over 40 partners to its ecosystem, jointly exploring the endless possibilities of AI.
During the event, MediaTek also unveiled its latest large language model for Traditional Chinese, MediaTek Research BreeXe. Built upon the Mixtral 8x7B model, it boasts significant advantages in resource efficiency and processing speed. According to official claims, MediaTek Research BreeXe outperforms GPT-3.5 in the TMMLU+ and MT Bench TW Traditional Chinese benchmark tests, establishing a new standard in AI language processing.
The introduction of the MediaTek DaVinci platform and the MR BreeXe model not only demonstrates MediaTek’s strong R&D capabilities in AI but also foreshadows an elevation in the level of AI-driven intelligence in Traditional Chinese applications. These innovative steps by MediaTek are set to propel global AI industry development, particularly advancements in Chinese language processing, ultimately providing users with smarter and more efficient interaction experiences.
【来源】https://www.ithome.com/0/760/772.htm
Views: 1