In the past year, there have been new trends in the development of artificial intelligence, and AI agents have become the most important direction for the development of large models Recently, Liang Zhixiang, Senior Vice President of Baidu Group, stated at the 13th China Intellectual Property Annual Conference that AI agents generated based on large models have truly achieved the transition from "human-computer interaction" to "human interaction". They can not only become "gold medal employees" for enterprises and "exclusive assistants" for users, but also build low threshold and easy distribution systems, allowing everyone to become a creator in the era of artificial intelligence.
Liang Zhixiang, Senior Vice President of Baidu Group
However, Liang Zhixiang also pointed out that the current battle of big models has just begun, and the gap in capabilities and costs is becoming larger and the competition will become more intense.
From "Human Computer Interaction" to "Everyone Dialogue"
Why are AI agents at the forefront of generative artificial intelligence? A McKinsey report shows that in the past year, Google, Microsoft, and OpenAI have all identified AI agents as the most important application scenarios for large models.
AI intelligent agents are driving a profound transformation from the 'human-computer interaction' mode to the 'human to human dialogue' mode. "Liang Zhixiang analyzed that AI intelligent agents have achieved deep learning and self evolution, and have been able to simulate more realistic ways of human dialogue.
With the understanding, logic, generation, and memory capabilities of large-scale models, AI agents can not only comprehend complex semantic relationships, but also perform intelligent reasoning and sentiment analysis based on context, making human-machine conversations as smooth and natural as conversations between people. AI agents can also instantly meet all user demands and even answer complex questions across domains.
In service scenarios, AI agents can be online at any time, achieving a 100% late night response rate. In the face of various professional questions from users, AI agents can also understand and not confuse. Taking a real story as an example, when a user encountered a difficult problem and initiated a consultation with an AI agent, they chatted for 5 hours and had 288 rounds of conversations. At present, the number of professional answers provided by merchant AI agents every day has exceeded 2.5 million.
In the product scenario, the large model also gives the intelligent agent an AI brain that can deeply interact with users. AI intelligent agents such as Huiboxing E-commerce Digital Person have achieved complete intelligence, unmanned operation, and full hosting of live streaming rooms.
The value of AI agents goes far beyond that. In today's society, data has become a new factor of production, and AI agents are key tools for mining the value of data and unleashing new quality productivity.
By processing and analyzing massive amounts of data through big models, AI agents can provide enterprises with more accurate market forecasts, customer insights, and decision support. This new quality of productivity not only improves the operational efficiency of enterprises, but also gives them an advantage in fierce market competition.
Cherries are delicious, and trees are also easy to plant
At present, the industry has clearly seen the development trend of the big model - from the virtual assistant to manage the schedule, to customer service, to the complex operation of autonomous vehicle, AI agents are participating in all aspects of life, helping to handle specific tasks or playing a role.
However, although AI agents are good, how to make users more willing to use and easier to use is a common problem faced in the application process of large models.
Regarding this, Liang Zhixiang gave a vivid analogy - "Cherries are delicious but difficult to plant," and the key to determining whether a tree is difficult to plant is "distribution".
If AI intelligent agents are like cherries, not only delicious but also easy to plant trees, allowing everyone to easily generate their own AI intelligent agents and accurately and efficiently distribute them to suitable users, can they truly unleash value. "Liang Zhixiang believes that only by lowering the threshold for AI native application development and distribution can more participants and innovators be attracted.
In the development of AI intelligent agents, the Wenxin Intelligent Agent Platform minimizes the development threshold and allows for the rapid construction of one's own AI intelligent agent applications without the need for new hardware or a large amount of additional training data. This low threshold feature has also enabled AI agents to be widely applied and promoted in various industries, truly achieving the goal of "developing AI agents in one sentence".
At present, Baidu Wenxin Intelligent Agent Platform has attracted 100000 enterprises and 600000 developers, covering hundreds of application scenarios. Whether individuals, enterprises, or institutions, they can create their own AI intelligent agents.
In terms of AI agent distribution, Baidu has formed an agent distribution pool represented by search. Developers only need to develop once to access Baidu's diverse and multi terminal scenarios such as search, text messaging, Baidu Xiaodu, Tieba, and online storage, thereby maximizing distribution effectiveness. At present, the distribution of AI agents in the Baidu ecosystem has significantly increased, with an average daily distribution exceeding 10 million in August.
Imagination is reality, and everyone is an AI agent developer, "said Liang Zhixiang as he looked forward to the future." New technologies are constantly emerging and emerging, and next year's big models will only develop even better and exceed everyone's imagination
Southern Metropolis Daily reporter Wang Chenchen