Written by 6:26 pm AI Security, Uncategorized

### Introducing Llama AI Model: Meta’s Latest Security Enhancements

Meta has launched “Purple Llama,” a project tied to its Llama AI tool, centering around safety tool…

A project related to Meta’s Llama artificial intelligence (AI) tool has been introduced as “Purple Llama.”

Announced on Thursday, December 6, Purple Llama focuses on safety equipment and assessment to aid developers in effectively developing their AI models.

Meta stated in its blog that the choice of lavender color signifies the adoption of both offensive (red team) and defensive (blue team) strategies to address the challenges posed by generative AI. Purple partnering involves a collaborative approach to identifying and mitigating potential obstacles, with responsibilities divided between the red and blue teams.

In July, Meta unveiled Llama 2 in partnership with Microsoft. The company revealed that over 100 million copies of its Llama designs have been preserved.

The project components are licensable “permissively,” allowing for research and commercial use to foster collaboration and standardize trust and security tools. Meta also mentioned that it is releasing what could be the first comprehensive set of large language model (LLM) cybersecurity safety assessments.

These benchmarks, developed in collaboration with safety experts, adhere to industry standards and guidelines. They aim to equip the White House with the necessary tools to address the risks outlined in its AI safety commitments.

Included in these tools are metrics for assessing the security risks associated with LLMs, techniques for monitoring the prevalence of concerning code suggestions, and mechanisms to impede the creation of malicious software or aid to cyber attackers. Meta believes that these tools will reduce the occurrence of worrisome AI-generated code suggestions from LLMs and their utility to malicious actors.

This announcement follows reports that Meta is testing Facebook, Instagram, Messenger, and WhatsApp across more than 20 conceptual AI features. These features encompass various aspects of social media such as exploration, advertising, and business communications.

Ahmad Al-Dahle, Meta’s Vice President of Relational AI, emphasized the company’s commitment to fostering a better community, enabling user expression, and delivering more valuable products. Al-Dahle noted that indicators of success include increased user engagement, positive feedback, and overall enjoyment among users.

Visited 2 times, 1 visit(s) today
Last modified: February 8, 2024
Close Search Window
Close