Meta Releases Code Llama: Empowering Developers in AI-assisted Coding

Meta, the parent company of Facebook, is expanding its suite of AI tools with the release of Code Llama. Building upon the success of its artificial intelligence model Llama 2, which enthralled the world last month, Code Llama is specifically designed for programming tasks. This release not only allows developers, startups, and researchers to experiment with advanced AI coding, but it also opens up new possibilities for embedding AI into software.

Deepak Kumar, a postdoctoral researcher at Stanford, highlights the significance of Meta’s move, expressing excitement that the weights of the neural network powering Code Llama are being made available to the community. Kumar notes that this open approach offers more flexibility for exploration compared to closed-source models from Google and OpenAI.

With Code Llama, developers have the potential to create innovative applications, such as programming assistants that provide additional safety checks before recommending code. Talia Ringer, an assistant professor at the University of Illinois Urbana-Champaign, anticipates Code Llama’s value for academic research. Additionally, varying assistant models specialized for specific coding needs could emerge.

However, to make sense of research on large language models like Code Llama, Ringer suggests that access to the training data is crucial. Meta claims that Code Llama is trained on publicly available code, outperforming existing open source coding models in two prominent benchmarks.

While GitHub’s Copilot, powered by OpenAI’s GPT, remains popular among developers, Code Llama offers a distinct advantage. Amjad Masad, CEO of coding platform Replit, suggests that Code Llama’s release opens doors for developers to experiment with agents that can perform useful tasks, such as browsing the web or utilizing APIs.

By positioning itself as a supplier of open AI tools, Meta aims to establish a presence in the competitive field of generative AI. Though Code Llama has certain usage restrictions, this strategic move by Meta may lock in researchers, generate new ideas, and potentially give the company an edge in the race to harness generative AI.

Sources:
– source article