According to Zhitong Finance App, Guosheng Securities released a research report stating that the intelligent body AutoGLM developed by the Zhifu technology team can simulate human operation of phones based on text/voice commands. In the long term, with the development of terminal AIAgent, there are significant growth opportunities for product demand and value for terminal manufacturers such as mobile phones, headphones, AR glasses. The terminal-side model is expected to promote the updating and replacement of terminals by demanding terminal hardware performance.

Zhongsheng Securities' main points are as follows:

Zhipu launches AutoGLM, and the intelligent assistant welcomes the moment of Phone Use.

On October 26, the Zhifu technology team introduced the intelligent body AutoGLM, which can simulate human phone operations with simple text/voice commands. The Android terminal currently supports apps such as WeChat, Taobao, Meituan, Dianping, Xiaohongshu, Gaode Maps, 12306, and Ctrip. During task execution, if there are multiple search results that meet the requirements or further user selection is required, AutoGLM will actively ask for the user's opinion. After the task is completed, users can provide feedback by dragging the floating ball left and right. Users can also create their own common commands. In the future, AutoGLM's capabilities will be more than that. Theoretically, with a deep understanding of GUI, AutoGLM can accomplish anything that humans can do on visual electronic devices, truly assisting humans in daily life and work.

Agent is an important direction for the industry, and Zhifu's technology level is leading.

1) On October 22, Anthropic announced the launch of the upgraded version Claude 3.5Sonnet, introducing the groundbreaking Computer Use feature. Developers can instruct Claude to use a computer like humans. AutoGLM also has Zhipu's Clear Speech plugin, which can perform web-based browser operations. OpenAI and Google are also laying out Agent technology. The bank believes that top companies in the industry are actively deploying Agents, clarifying the industry direction. Shortly after Claude introduced "Computer Use," Zhipu launched "Phone Use," verifying that China's large-scale model companies are at the forefront of the world in terms of technological capabilities.

From a technical perspective, AutoGLM is based on the self-developed 'Basic Intelligent Decoupling Intermediate Interface' and 'Self-evolution Online Course Reinforcement Learning Framework' by Zhipu, addressing two key challenges of large models as intelligent agents - insufficient precision in action execution and lack of flexibility in task planning. Compared to computers, smartphones currently accompany most users for a longer time. If 'Computer Use' has opened up a new paradigm of human-computer interaction, then 'Phone Use' takes it a step further, unlocking more possibilities for applications, truly benefiting everyone with AI.

Future Outlook for Intelligent Assistants: Terminal manufacturers and inference-side computing power are expected to greatly benefit.

The bank believes that in the long term, mainstream smartphone manufacturers are highly likely to launch their own Agent applications through self-research or cooperation with large model manufacturers. AutoGLM requires mobile operating systems to grant accessibility, floating window, microphone, and screen recording permissions. Terminal hardware manufacturers control the flow entry, potentially having more say in the future AI terminal ecosystem compared to app manufacturers. In addition to smartphones, other terminal intelligent assistants such as headphones continue to evolve, such as ByteDance's release of the AI smart assistant earphone OlaFriend.

The bank believes that benefiting from the development of edge-side AIAgents, there are significant growth opportunities for products and value from smartphone, headphone, AR glasses, and other terminal manufacturers. As various AIAgents are widely used, the demand for inference-side computing power is expected to increase significantly. In the future, with the enhancement of terminal computing power and continuous model optimization, AIAgent models are expected to be widely deployed at the edge. The requirement of edge-side models for terminal hardware performance is expected to drive terminal upgrades.

Recommendations:

Potential Hardware: Edifier Technology (002351.SZ), Shenzhen Transsion Holdings Co., Ltd. (688036.SH), Thunder Software Technology (300496.SZ);

Zhipu's Partner Companies: Dou Shen Education (300010.SZ), Hunan TV & Broadcast Intermediary (000917.SZ);

Supply Chain: Luxshare Precision Industry (002475.SZ), Suzhou Dongshan Precision Manufacturing (002384.SZ), Avary Holding (002938.SZ);

Computing power: Cambricon Technologies (688256.SH), dawning information industry (603019.SH), inspur electronic information industry (688041.SH), inesa intelligent tech inc. (600602.SH), surfilter network tech (301236.SZ), zhongji innolight (300308.SZ), eoptolink technology inc. (300502.SZ), inspur electronic information industry (000977.SZ), foxconn industrial internet (601138.SH), digital china group (000034.SZ), sharetronic data technology (300857.SZ), xiamen hongxin electronics technology group inc. (300657.SZ), netac technology (300042.SZ), inesa intelligent tech inc. (688343.SH), chengdu hi-tech development (000628.SZ), etc.

Risk Warning: Risks of AI technology iteration falling short of expectations; risks of economic downturn exceeding expectations; risks of intensified industry competition.