ERNIE AGENT

Introduction

The ERNIE agent technology is inspired by the human brain's fast and slow thinking processes. The ERNIE agent is built based on the thinking model trained on the foundation model enhanced with thinking training, and has equipped with capabilities of understanding, planning, rethinking and evolving. The ERNIE agent technology augments ERNIE foundation model's capabilities and accelerates the explosion of Al applications.

Innovative Agent Technology Based on Thinking Model

The ERNIE agent technology, inspired by the human brain's fast and slow thinking processes, has further developed a slow thinking mechanism based on a thinking model trained on the foundational model, making it more rational and accurate. The thinking model, through thinking-enhanced training such as supervised fine-tuning of the thinking process, preference learning in behavior decision-making, and reinforcement learning through output rethinking, possesses more powerful capabilities in understanding, planning, rethinking and evolving. It ensures reliable execution and autonomous evolution, allowing machines to think and act like humans, autonomously completing complex tasks while continuously learning and evolving in the environment.

The breakthrough in ERNIE agent technology has significantly enhanced the capabilities of ERNIE foundation model. According to authoritative evaluations by IDC, Frost & Sullivan and the China Software Testing Center, ERNIE 4.0's performance exceeds the average performance of international foundation models, ranking first in China and among the top globally.

Facilitating the Burst of Al Applications, Empowering Thousands of Industries

The ERNIE Agents are applied in the fields of intelligent assistant, customer service, enterprise intelligence, etc., fully unleashing the potential of large models, promoting the large-scale industrial application of Al. By the end of September, on the New APP, 300,000 creators had created 400,000 agents, equipped with multiple functions. Agents' daily request reached a total of 800 million times.

The Agent Builder is built based on agent technology to further reduce the threshold of development and eliminate the last barrier of agent application from development to distribution. Developers can use diversified capabilities and tools based on their own industry fields and application scenarios to create Al-native applications in the era of large models. The platform provides developers with a distribution path of Baidu's ecosystem matrix. By the end of September, it had attracted 600,000 developers and 100,000 enterprises, with an average daily distribution of over 8 million.

Reshaping Work and Life, Accelerating Industrial Upgrading

Agent technology is driving the deep integration of large model technology into various sectors in society, thereby empowering the upgrading of businesses and social services to be more intelligent. In terms of government services, Al-powered legal consultants offer workers convenient and effective legal services, bolstering public trust and satisfaction with the government. All in all, government services become more intelligent thanks to agent technology. In agriculture, the Farmer Academician Agent can teach farmers planting knowledge and skills anytime, anywhere, making it more convenient for technology to assist and benefit farmers. In education, agent technology offers teachers and students a more intelligent and personalized teaching and learning experience respectively. Furthermore, agent technology can be used to help programming, not only lowering the threshold for ordinary people to develop applications, but also decreasing the workload of professional programmers, and substantially enhancing the efficiency of human-computer interaction.

The traditional industries have achieved innovation in production methods and management modes, relying on the widespread application of agent technology. At the same time, emerging industries such as smart home and smart e-commerce have rapidly risen, and also give birth to new job opportunities, such as agent engineers, agent trainers, agent evaluators, and many new positions for the production, maintenance, and supervision of agents. These new positions will not only enrich people's career choices, but will also promote innovation in talent cultivation modes.