Amazon on Thursday unveiled CodeWhisperer, a new tool for developers that denerates code recommendations. The machine learning-powered tool is currently available in preview.
CodeWhiperer provides recommendations based on contextual information, such as the cursor location in the source code, code that precedes the cursor, and code in other files in the same project. It also responds to simple natural language prompts, such as “upload a file with server side encryption.”
CodeWhisperer leverages the latest in large language models and is trained on huge datasets – open source repositories, internal Amazon repositories, API documentation and forums, etc.
“We trained the model on the most common patterns for building cloud applications, so you can build and innovate on the cloud much faster than ever before,” Swami Sivasubramanian, AWS VP of data and ML services, said during the Amazon re: MARS conference on Thursday.
While it’s designed to boost coder productivity, it does so in a responsible manner, he added – mitigating risks like bias, security vulnerabilities, and bugs. It comes with a built-in security scanner to help detect vulnerabilities in developers’ projects. It also has a built-in reference tracking feature to detect whether a code recommendation may be similar to particular CodeWhisperer training data. This allows you to easily find and review that reference code and explore how it’s used in the context of another project.
CodeWhisperer will also help coders avoid bias by removing code recommendations that may be considered biased and unfair.