TL;DR:
- Top AI developers, including OpenAI, Google, and Nvidia, are collaborating for a public evaluation of their generative AI systems at DEF CON 31 in Las Vegas.
- The event will be hosted by AI Village, a community of AI hackers.
- The White House endorses pushing generative AI models to their limits to provide critical information to researchers and the public about their impacts and enable developers to fix any issues found.
- The AI model assessment will use an evaluation platform developed by Scale AI and will be the largest red teaming exercise ever for any group of AI models.
- Participants will have timed access to multiple LLMs through laptops provided by the organizers, and the person with the most points will win a high-end Nvidia GPU.
- The aim of the exercise is to grow the community of researchers equipped to handle vulnerabilities in AI systems and improve overall security and resilience.
Main AI News:
The White House has announced an unexpected collaboration between some of the biggest names in the AI industry, including OpenAI, Google, Antrhopic, Hugging Face, Microsoft, Nvidia and Stability AI. The collaboration is set to participate in a public evaluation of their generative AI systems at DEF CON 31, a hacker convention scheduled to take place in Las Vegas in August. The event will be hosted by AI Village, a community of AI hackers.
The popularity of large language models (LLMs) such as ChatGPT has increased significantly over the past year, with people using them to accelerate writing and communication tasks. However, officials acknowledge that these models come with inherent risks, including confabulations, jailbreaks, and biases, which pose challenges for security professionals and the public. That’s why the White House Office of Science, Technology, and Policy supports pushing these new generative AI models to their limits.
“The independent exercise will provide critical information to researchers and the public about the impacts of these models and will enable AI companies and developers to take steps to fix issues found in those models,” said a statement from the White House. The event aligns with the Biden administration’s AI Bill of Rights and the National Institute of Standards and Technology’s AI Risk Management Framework.
According to AI Village organizers Sven Cattell, Rumman Chowdhury, and Austin Carson, the upcoming event will be “the largest red teaming exercise ever for any group of AI models.” Thousands of people are expected to participate in the public AI model assessment, which will use an evaluation platform developed by Scale AI.
“Red-teaming” is a process in which security experts attempt to find vulnerabilities or flaws in an organization’s systems to improve overall security and resilience. By conducting the largest red-teaming exercise for any group of AI models, AI Village and DEF CON aim to grow the community of researchers equipped to handle vulnerabilities in AI systems.
LLMs have proven to be difficult to secure, in part due to a technique called “prompt injection,” which can derail a language model into performing unintended actions. During the DEF CON event, participants will have timed access to multiple LLMs through laptops provided by the organizers. A capture-the-flag-style point system will encourage testing a wide range of potential harms. The person with the most points at the end will win a high-end Nvidia GPU.
“We’ll publish what we learn from this event to help others who want to try the same thing,” wrote AI Village. “The more people who know how to best work with these models, and their limitations, the better.”
DEF CON 31 is set to take place on August 10–13, 2023, at Caesar’s Forum in Las Vegas. Stay tuned for updates on this exciting event!
Conlcusion:
The public evaluation of generative AI systems at DEF CON 31 represents a significant development for the AI industry. This collaboration between major players in the field, such as OpenAI, Google, and Nvidia, provides an opportunity to assess the risks associated with large language models and identify ways to address them.
This exercise will help to build a community of researchers equipped to handle vulnerabilities in AI systems, which will ultimately improve the security and resilience of these systems. As the market for AI continues to grow, events like DEF CON 31 are essential to ensure that the technology is developed and deployed in a responsible and secure manner.
