In a move to foster ethical use of artificial intelligence (AI), Meta Platforms Inc. META has introduced “Purple Llama,” a project designed to help developers create responsible AI, the company announced Wednesday.
Meta unveiled that the project is designed to amalgamate tools and evaluations to aid developers in responsibly developing open generative AI models. The project takes its unique color name from cybersecurity’s concept of purple teaming, which combines offensive and defensive strategies, offering a comprehensive approach to assessing and mitigating potential risks.
The initial version of “Purple Llama” will feature tools and evaluations for cybersecurity and input/output safeguards. These components will be permissively licensed, making them accessible for both research and commercial purposes.
As a part of its commitment to open and transparent science, Meta is also launching “Llama Guard,” a fundamental model to help developers circumvent the generation of potentially risky outputs. This model has been trained on various publicly available datasets to detect commonly risky or violating content.
Meta envisions enabling developers to tailor future versions based on their specific requirements, thereby making it easier to adopt best practices and enhance the open ecosystem. The company looks forward to collaborating with partners who share a similar vision for an open ecosystem of responsibly-developed generative AI.
Why It Matters: The launch of Purple Llama represents Meta’s ongoing efforts towards safe AI development, following LLaMA’s launch in February this year to assist researchers and engineers in advancing AI applications.
Furthermore, the company’s emphasis on AI was highlighted during its annual Connect conference, where AI took center stage, indicating a shift in the company’s focus from virtual reality to AI development. Competing with OpenAI‘s ChatGPT, Microsoft’s Bing Chat, some AI researchers believe that LLaMa will reshape the LLM landscape, despite many investors finding its value proposition elusive.
Image Credit – Shutterstock
Engineered by
Benzinga Neuro, Edited by
Pooja Rajkumari
The GPT-4-based Benzinga Neuro content generation system exploits the
extensive Benzinga Ecosystem, including native data, APIs, and more to
create comprehensive and timely stories for you.
Learn more.
Image and article originally from www.benzinga.com. Read the original article here.