Formerly known as Facebook, Meta has announced the launch of its new, free, state-of-the-art AI tool that can process text prompts that streamline the process of code creation and debugging, according to a company press release on Aug. 24. .
Dubbed ‘Code Llama’, the tool uses Meta’s Llama 2 major language model (LLM) to make workflows faster and more efficient for developers and beginners who want to learn how to code.
Designed as a free educational resource to help programmers write “more robust, well-documented software,” Code Llama can generate code and natural language about the code from both the code itself and natural language prompts such as “write me a function that performs the “Fibonacci Sequence”. It can also reportedly be used for commercial purposes free of charge.
One of its most notable features is its adaptability to different scenarios: it can assemble segments of code from these text prompts, or identify and correct (ie debug) problems within specific code strings.
According to Meta, Code Llama stems from the growing trend of programmers using LLMs to improve various aspects of software development, ranging from creating new applications to troubleshooting existing code bases.
Meta’s strategy is to release Code Llama in three different sizes to meet different project requirements. Notably, even the smallest variant is designed to run on a single Graphics Processing Unit (GPU), making it ideal for projects requiring low latency responsiveness.
Other code generators
In addition to Meta’s code generator tool, we also see other types of code generators emerging. However, the problem of copyright infringement remains a major setback.
Last November, there was a class action lawsuit against Github’s parent company Microsoft, as well as OpenAI, over allegations of copyright infringement involving GitHub’s “Copilot” tool powered by OpenAI’s GPT-4.
Google’s DeepMind team has also teased its AlphaCode coding tool, but has not yet released it to the general public. The company is also currently testing an internal AI tool that could give people life advice and perform at least 21 different tasks.
Similarly, Amazon Web Services (AWS) uses ‘CodeWhisperer’ to write code and debug errors.
As for Meta, the company has gone a step further by releasing specialized iterations of Code Llama. One such version is Code Llama-Python, tailored for Python programming, and another is Code Llama-Instrct, designed to understand natural language instructions. It is important to note that these different iterations are not interchangeable. The company strongly discourages using the Llama base code or Code Llama-Python for interpreting natural language instructions.
That said, while Meta’s latest innovation positions it as a formidable competitor, conversations about ethical use will continue to surface as the AI landscape continues to evolve.