GitHub has updated its Copilot coding assistant with new features, such as a “vulnerability filtering system” to prevent insecure coding patterns such as SQL injection and hard-coded credentials.
The level of acceptance of suggested code has also increased thanks to improved AI models and techniques, rising from 27% in June 2022 to 46% for Java code. Improved methods for context understanding, an updated client-side model that fewer unwanted suggestions, and an updated OpenAI Codex model are some of the new features.
According to Senior Director of Product Management Shuyin Zhao’s post, the vulnerability scanner approximates the behavior of static analysis tools by using LLMs (Large Language Models).
Read More: GitHub claims new smarter Copilot will block insecure code, writes 40-60% of developer output