Meta has introduced Purple Llama, a project dedicated to creating open-source tools for developers to evaluate and improve the reliability and safety of generative artificial intelligence (AI) models before they are put into use.
Meta stressed the need for collaborative efforts in ensuring AI safety, arguing that AI challenges cannot be solved in isolation.
The company says the goal of Purple Llama is to establish a common platform for developing safer generative AI as concerns grow about large language models and other AI technologies.
Gareth Lindahl-Wise, chief information security officer at cybersecurity firm Ontinue, calls Purple Llama “a positive and proactive step” toward safer AI technology.
The Purple Llama project partners with AI developers; cloud services like AWS and Google Cloud; semiconductor companies like Intel, AMD, and Nvidia; and software companies, including Microsoft.
The first toolkit released through Purple Llama is CyberSecEval — a tool for assessing cybersecurity risks in AI-generated software.
It has a language model that identifies inappropriate or harmful text, including discussions of violence or illegal activity.
Developers can use CyberSecEval to test whether their AI models are prone to generating unsafe code or supporting cyberattacks.
Meta research has found that large language models often suggest vulnerable code, highlighting the importance of continuous testing and improvement for AI security.
Llama Guard is another toolkit — a large language model trained to identify potentially harmful or offensive language.
Developers can use Llama Guard to check whether their model generates or accepts unsafe content, helping to filter out prompts that may lead to inappropriate output.
(according to IFW)
UK invests 'hugely' in AI, launches 5 new quantum projects
AI GUIDi Belt Replaces Canes for the Visually Impaired
Tech giants' open-source AI alliance will compete with OpenAI
A unique direction in AI chatbot development in China
Source
Comment (0)