RockAI CMO 邹佳思：端侧智能如何通过「原生记忆」与「自主学习」，完成从工具迈向伙伴的人机关系丨GAIR 2025

Core Viewpoint - The article discusses the potential of edge intelligence as an alternative path for AI development, especially as the limitations of Transformer models become apparent [1]. Group 1: Conference Overview - The 8th GAIR Global Artificial Intelligence and Robotics Conference was held in Shenzhen, focusing on AI's evolution and its impact on various sectors [2][3]. - The conference featured notable speakers, including CMO of RockAI, who emphasized the need to move beyond the constraints of Transformer models [3]. Group 2: Edge Intelligence Concept - Edge intelligence allows for local deployment of AI models, enabling devices to operate without cloud involvement, thus enhancing privacy and reducing costs [4][9]. - The current cloud model, which relies on token payments, is criticized for being inefficient, with over 50% of token consumption deemed wasteful [4][9]. Group 3: Challenges and Innovations - Transitioning to edge intelligence faces challenges such as limited computational resources and the need for devices to possess learning capabilities [13][15]. - RockAI aims to develop non-Transformer models that incorporate native memory and autonomous learning, fostering a "collective intelligence" ecosystem [4][23]. Group 4: Future Directions - The future of AI hardware should focus on real-time learning and adaptability, moving away from static knowledge bases [21][19]. - The development of RockAI's Yan model, which integrates memory modules and selective activation mechanisms, represents a significant step towards achieving these goals [23][31]. Group 5: Practical Applications - Edge models can facilitate complex interactions between devices, enhancing user experience in everyday scenarios, such as smart home automation [27][29]. - The integration of edge intelligence in consumer electronics is expected to lead to more personalized and emotionally aware devices [29][31]. Group 6: Collective Intelligence - The concept of collective intelligence suggests that interconnected devices can collaborate to solve problems, similar to human cooperation [33][35]. - The article posits that as the limitations of large-scale models become evident, innovation in architecture is necessary to avoid stagnation in AI development [35].