Core Viewpoint - The founder of 360 Group, Zhou Hongyi, emphasized the need for new security measures to address challenges posed by AI, proposing the concept of "using large models to combat large models" [1] Group 1 - Zhou Hongyi introduced the idea of a "large model guardian" to tackle risks such as attacks on large models, privacy breaches, and content errors [1] - The proposed solution aims to alleviate issues related to hallucinations in AI systems [1]

360集团创始人周鸿祎谈到，风险还是很多的。他举例到，比如说做一个常规的it系统，它肯定会有漏 洞、有数据隐私的泄露问题，但他提到，这些传统安全都不足为虑，最要命的是有三个安全问题： 他提到，大模型降低了每个人的使用门槛，没学过编程的人现在通过跟大模型交流也能写程序，这就使 得大模型被攻击的门槛也降得很低。通过精心构造了指令，可以让大模型完全听从指令，比如说可以把 企业的机密文件交给你，这种事情已经发生了，这叫做"注入攻击"。"甚至我有时候开玩笑说，一个前 台小姐姐可能不会编程，她对老板不满意，她都可以对大公司的大模型和智能体进行攻击。" 第一个问题，大模型会有错觉，会胡说八道，会出错，甚至还有幻觉。 周鸿祎表示，当大模型有些事情不明白的时候，就会一本正经的瞎编，这个在大模型被当作玩具的时候 可以一笑而过，但是今天大模型带着智能体将要进到很多工业生产制造产品，以及很多政府部门的办公 领域，这种出错可能就是比较可怕的。他强调，特别是有了智能体之后，智能体又有能力去操纵各种工 具，这个危害影响就会更大。 第二个问题，大模型使得任何人攻击人工智能变成可能性。 专题：2025中国互联网大会 7月23日上午消息，202 ...

Group 1 - The application of AI presents a rare opportunity for traditional manufacturing industries, emphasizing the need for scenario-driven applications [1][3] - AI enhances user experience and requires companies to rapidly adapt their thinking and respond to market demands, particularly in decision-making processes [3][5] - AI is expected to replace repetitive jobs while simultaneously creating new positions, necessitating societal adjustments in training and job transitions [5][6] Group 2 - The development of AI and robotics is driven by market demand and application scenarios, with companies like Haier already investing in both industrial and household robots [9][11] - Continuous progress and adaptation to technological changes are crucial for companies in the robotics sector to remain competitive [11] - The importance of establishing supportive policies and ethical regulations before implementing AI in business projects was highlighted, with a call for employee involvement in standard discussions [8]