AI in practice

Meta's Purple Llama aims to make open-source AI safer

Matthias Bastian
A portrait of a normal llama wearing a full-body purple raincoat. The raincoat is sleek and shiny, covering the llama from head to toe in a bright purple color. It features a hood that gently covers the llama's head and large buttons down the front. The background is a simple, outdoor setting, possibly hinting at rainy weather, enhancing the appropriateness of the raincoat. The image captures the llama standing calmly, showcasing the unique outfit in a natural, serene setting.

DALL-E 3 prompted by THE DECODER

Meta AI has launched Purple Llama, an umbrella project aimed at providing open trust and safety tools for responsible generative AI development and "to level the playing field." The project will offer tools and evaluations to help developers build responsibly with open generative AI models, with initial focus on cybersecurity and input/output safeguards. As part of the project, Meta AI is releasing CyberSec Eval, a set of cybersecurity safety evaluation benchmarks for large language models (LLMs), and Llama Guard, a safety classifier for input/output filtering. Purple Llama is supported by partners such as AI Alliance, AMD, AWS, Google Cloud, Hugging Face, IBM, Intel, Microsoft, MLCommons, NVIDIA, and Scale AI, among others.

Image: Meta AI
Sources:
Meta