At the CCS2024 cybersecurity event held in Chengdu, Baidu's Vice President Chen Yang delivered a keynote speech on how large-scale model technology is driving transformation in the software development field. Chen Yang highlighted that the integration of large models' capabilities in understanding, generating, logic, and memory with software research and development can significantly enhance development efficiency and improve software security.
Baidu has been exploring the field of intelligent coding for two years, and its smart coding efficiency tool, "Ernie Quick Coding," has been widely adopted within the company, significantly boosting engineers' R&D efficiency. Currently, 30% of the code generated daily at Baidu is produced by Ernie Quick Coding, with an overall adoption rate of 46%, and a 12% efficiency improvement for engineers. This not only increases the number of code submissions and the speed of business iteration but also allows engineers to focus on more valuable and creative work.
Chen Yang emphasized that while Ernie Quick Coding enhances R&D efficiency, it also places a high priority on engineering quality. The tool integrates quality and security considerations throughout the entire R&D process, including requirement clarification, security design, module sorting, code explanation, security vulnerability scanning, and repair, ensuring the security and quality of the code. In Baidu's internal practices, the accuracy rate of security vulnerability scanning by Ernie Quick Coding exceeds 95%, with 83% of the scanned vulnerabilities being fixed.
Although large-scale model technology has demonstrated significant value and practical effects in the field of intelligent R&D, challenges still exist in the application of private domain knowledge, differences in industry quality standards, engineering culture coordination, and value measurement. Baidu, through the internal and external experiences of Ernie Quick Coding, has established a standardized implementation process and best practices, promoting a shift in corporate engineering culture and the construction of a data-driven value cycle.