(Credit: Shutterstock/Rafapress)
Google is launching two new AI coding tools based on its Gemini and Gemma AI models.
The first is Gemini Code Assist, which is based on Google's Gemini model, formerly known as Bard. Gemini Code Assist is designed to write code in C, C++, MatLab, Ruby, Rust, JavaScript, Python, and SQL, to name a few. It can currently handle 22 different programming languages and can also handle questions about Google Cloud, debug existing code, or explain your own code to you in more detail. Code Assist is part of the company's Gemini for Google Cloud offering, which Google is selling to businesses.
Dubbed "CodeGemma," Google's new Gemma AI variant is able to code in Python, JavaScript, and Java, to name a few mentioned in Google's announcement. The tech giant is releasing three different versions of CodeGemma. One "7B" version is designed for code generation and existing code line completion, while a second 7B version is more tuned for "code chat and instruction-following." A third smaller, "2B" version is one intended to be downloaded locally to one's computer and can fulfill code completion requests quickly, Google says.
The CodeGemma models were trained on 500 billion tokens of data including "web documents, mathematics, and code." Google's post suggests that CodeGemma could be a good tool to help engineers reduce the amount of generic "boilerplate" code they have to write.
Two original general-purpose versions of Gemma were first released back in February. Gemma's base models aren't currently available to the general public, but can be accessed through Nvidia's Chat With RTX or the Opera One browser's developer version.
Google's Gemini Code Assist and CodeGemma are the latest additions to the growing number of AI coding tools developers can use to complete projects. Previously, Google offered Duet AI as a coding tool. Gemini Code Assist could be an alternative to Microsoft's GitHub Copilot, which can also write and complete code. But Google isn't offering a straightforward subscription option for individuals like Copilot, which charges solo developers $10 a month for access.
CodeGemma, Gemini Code Assist, and GitHub Copilot are intended to work in tandem with developers, but there's also Devin, an "AI software engineer" that promises to complete entire engineering projects with little to no guidance. Devin, which uses OpenAI's Chat GPT-4 Turbo, was announced last month from Cognition Labs. The startup claims that Devin can train other AI models and code entire websites by itself, among other tasks.
While some of these AI tools could lead to fewer software engineering jobs, it's also possible the AI boom could create more jobs working on AI models in the short-term. Or, AI tools could help developers complete more tasks faster—as long as the AIs don't hallucinate too much.