字节跳动豆包公关负责人刘星就国内算力配置问题发文回应1。针对"国内超一半算力在造爽剧"的说法，刘星否认了相关数据的准确性1。他表示，在字节总算力中，只有约10%用于视频生成模型的训练和推理1，与市场传言存在较大出入1。
此前有投资者关系活动记录表披露，国内AI Token消耗中，55%烧在短剧和视频生成，24%烧在电商广告，15%用于研发1。该数据还显示，AI短剧与视频生成的Token消耗量已达到软件开发的两倍以上1。
Liu Xing, head of public relations for ByteDance's Douyin, has refuted assertions that the majority of China's artificial intelligence computing capacity is being consumed by short drama and video generation.1 According to Liu's statement, only approximately 10% of ByteDance's total computing power is allocated to training and inference for video generation models.1 This contradicts market claims suggesting that 55% of domestic AI token consumption is being directed toward short dramas and video generation.1
The disputed figures originate from investor relations documentation disclosed by China Online, which indicated that 55% of national AI token consumption is being spent on short dramas and video generation, with an additional 24% directed toward e-commerce advertising and 15% dedicated to research and development.1 According to the same data, token consumption for AI short dramas and video generation has reached more than double that of software development.1
评论
还没有评论，欢迎留下第一条。