社交网络里面每一个人都有一个数字身份。Second me要做的就是这个身份模型。 "我就是想把Facebook重新做一遍。"前Meta员工、心识宇宙创始人陶芳波开门见山地对凤凰网科技表示。 在几乎所有AI创业者都在追逐"Agent（智能体）"浪潮的当下，他却选择了一条更激进、也更浪漫的路：用AI重塑人的数字身份，构建一个以"AI身份"为 节点的下一代社交网络。 今年外滩大会期间，凤凰网科技与陶芳波有过一次深入交流，陶芳波提出的"身份模型"，即通过开源项目"Second Me"实现个人AI身份，让AI成为对抗超 级智能的"第二自我"。 陶芳波曾在美国微软研究院、NASA 、Facebook Research 等企业与机构从事 AI 研发工作，并在阿里巴巴达摩院创建了神经符号实验室。2022年创立 Mindverse 心识宇宙，获红杉、线性、Square Peg 等亿级融资。 2024年，心识宇宙推出了个人AI助手Me.bot，要做每一个人的"AI身份"。2025年，心识宇宙开源了Me.bot的核心技术"Second Me"，用户可以下载到本 地，创造出代表自己的Agent，是今年 GitHub 平台上关注度增速 ...

Core Viewpoint - The article discusses the innovative approach of Mindverse in creating AI identity models that reflect users' preferences, values, and memories, aiming to enhance user interaction with AI and digital platforms [6][7][8]. Group 1: AI Identity Models - Mindverse is developing a third type of AI assistant, termed "identity model," which aims to replicate a user's "second self" by fine-tuning a base language model with user-specific data [6][8]. - The identity model is designed to understand and represent users in various applications, allowing for more personalized interactions with AI [7][8]. - The company believes that having an identity model can significantly improve efficiency in online interactions, as it can autonomously initiate tasks and manage communications [8][11]. Group 2: User Engagement and Product Development - The app Me.bot, launched in May 2022, has attracted nearly one million users and is designed to help users develop their AI identity models through daily interactions [8][11]. - Mindverse has initiated an open-source project called Second Me, which has gained significant traction on GitHub, indicating strong community interest in the identity model training methods [8][33]. - The company emphasizes the importance of user engagement by integrating the AI into daily life, encouraging users to record their experiences and thoughts [20][21]. Group 3: Technological Insights - The approach to training identity models is inspired by human cognitive processes, where the model learns to index and connect relevant information rather than storing fragmented knowledge [7][22]. - Mindverse's identity models are trained daily to keep up with users' evolving experiences and self-perceptions, mirroring how human memory works [29][30]. - The cost of training an identity model is relatively low, with estimates around one dollar per training session for a model with 7 billion parameters [30]. Group 4: Market Potential and Future Directions - The identity model can potentially replace traditional user interactions with digital platforms, allowing for more seamless and efficient communication [31][37]. - Mindverse is exploring monetization strategies, including charging users for identity services and collaborating with platforms to understand user preferences [36][37]. - The company anticipates that as AI technology matures, the integration of identity models into existing digital ecosystems will become more prevalent, enhancing user experience [32][36].