摘要
文生视频大模型Sora超越了人工智能的文本表达结构,不断强化图像与视频的价值意义,其话语建构能力日益趋近人类的创作水平。这种“超能力”背后隐喻着“技术”的本质。生成式人工智能的技术机理主要通过资源堆叠、机器学习与序列预测来实现,其中资源堆叠奠定了智能根基、机器学习促进了知识萃取、序列预测形塑内容体系。世界模拟想象和人机关系演进共同构筑了生成式人工智能的意义价值。然而,随着生成式人工智能的普泛化渗透,其遮蔽的想象塌缩、资源消耗以及内容速朽等困境日渐凸显,亟须通过价值对齐、黑箱稀释等实践方式回应人机协同进化的时代命题。
Sora,as a text-to-video large AI model,transcends the conventional text expression structures of artificial intelligence and continually enhances the value and significance of images and videos.Its discourse construction capabilities are increasingly approaching the level of human creativity.Behind this"Superpower"lies a metaphor for the essence of technology.The technical mechanisms of generative artificial intelligence are primarily realized through resource stacking,machine learning,and sequence prediction.Resource stacking establishes the foundation of intelligence,while machine learning facilitates knowledge extraction,and sequence prediction shapes the content system.The imaginative simulation of the world and the evolution of the humanmachine relationship jointly construct the significance of generative artificial intelligence.However,with the widespread adoption and integration of generative artificial intelligence,the challenges of hidden imaginative collapse,resource consumption,and content decay have become increasingly prominent.It is imperative to address the contemporary challenge of human-machine co-evolution through practical methods such as value alignment and black box dilution.
出处
《决策与信息》
2024年第9期87-96,共10页
Decision & Information
关键词
大模型
生成式AI
机器学习
SORA
人工智能
人机关系
AI技术-社会协同
Large model
generative AI
machine learning
Sora
artificial intelligence
human-machine relationship
AI Technology-Social Collaboration
