大模型如何更懂“中国话”（“十五五”文化热词·推进文化和科技融合）

Core Viewpoint - The article discusses the significance of increasing the proportion of Chinese data in training large AI models, emphasizing its impact on understanding Chinese language nuances, cultural context, and enhancing the model's performance in various applications [1][2][3]. Group 1: Importance of Chinese Data - The increase in Chinese data proportion in AI model training enhances the model's understanding of unique Chinese cultural expressions and terminologies, which are often not represented in English data [2][3]. - High-quality Chinese data is crucial for the accurate representation of traditional knowledge, such as Chinese medicine, which requires contextual understanding that English data cannot provide [3][4]. - The reliance on English data poses risks for Chinese AI models, as it may lead to biases and inaccuracies in understanding Chinese-specific contexts and terminologies [2][3]. Group 2: Quality of Chinese Data Supply - There is a distinction between ordinary Chinese data and high-quality Chinese data, with the latter being verified and sourced accurately, which is essential for reliable AI outputs [4][5]. - The supply of high-quality Chinese data is improving due to supportive policies, technological advancements, and a growing consensus in the industry regarding the need for Chinese-adapted models [5][6]. - The development of high-quality data sets is being accelerated by initiatives such as the establishment of data annotation bases and the implementation of new technologies to reduce costs and improve efficiency [5][6]. Group 3: Challenges and Recommendations - The current landscape of Chinese data is characterized by redundancy and a lack of high-quality content, particularly in specialized fields like healthcare, necessitating the establishment of standardized data quality metrics [7][8]. - There is a need for technological solutions to overcome data silos and compliance issues, enabling collaborative data annotation across institutions while ensuring privacy [7][8]. - The demand for Chinese data in emerging fields, such as the metaverse and traditional cultural practices, highlights the necessity for targeted data collection efforts to enhance digital resources [8][9]. Group 4: Cultural and Technological Integration - The integration of culture and technology is emphasized as a means to enhance cultural production processes and promote digital transformation in cultural sectors [9]. - The article suggests that AI technologies can enrich cultural expressions and create new consumption scenarios, thereby fostering innovation in cultural industries [9].