Meta announces Purple Llama initiative to promote responsible and safe generative AI development

Meta Platforms Inc. is looking to promote the responsible development of artificial intelligence systems with a new initiative called Purple Llama. Announced today, it’s a project that Meta says will slowly make available all of the tools and evaluations the AI development community needs to build open generative AI models in a safe and responsible way. In a blog post, Meta’s AI research team explains that generative AI is powering a wave of innovation, with conversational chatbots, image generators, document summarization tools and more now being widely used across the world. Many of these new generative AI applications are powered by Meta’s open-source Llama large language models. As such, Meta feels compelled to take responsibility and encourage collaboration on AI safety, saying this is necessary to build trust in these new innovations. “The people building AI systems cannot address the challenges of AI in a vacuum, which is why we want to level the playing field and create a center of mass for open trust and safety,” the company explained. To get things started, Meta is kicking off Purple Llama with the release of a free and open set of cybersecurity evaluation benchmarks for LLMs, called CyberSec Eval. It’s also announcing Llama Guard, which is a safety classifier for input/output filtering that has been optimized for easy deployment. In a blog post, Meta says cybersecurity and LLM prompt safety are two of the most important areas in generative AI safety, as highlighted within its Llama 2 Responsible Use guidelines. CyberSec Eval is believed to be the first industrywide set of cybersecurity safety evaluations for LLMs. The benchmarks are based on industry standards and guidance, and were developed in collaboration with security subject matter experts, Meta said.

Full story : Meta announces Purple Llama, an initiative to promote responsible AI development through the release of tools and evaluations to build open generative AI models.