New Top AI Developers Join Forces for Public Evaluation of Language Models at DEF CONPost

The White House has announced an unprecedented collaboration between top AI developers, including OpenAI, Google, Antrhopic, Hugging Face, Microsoft, Nvidia, and Stability AI, to participate in a public evaluation of their generative AI systems at DEF CON 31, a hacker convention to be held in Las Vegas in August. This exercise aims to push the limits of large language models (LLMs), such as ChatGPT, which have become increasingly popular in recent times. However, the White House recognizes that these models come with inherent risks such as confabulations, jailbreaks, and biases, which pose significant challenges for security professionals and the public. Therefore, they endorse the exercise, which aligns with the Biden administration's AI Bill of Rights and the National Institute of Standards and Technology's AI Risk Management Framework.

The event will be hosted by AI Village, a community of AI hackers, and thousands of people will participate in the public AI model assessment. The evaluation platform developed by Scale AI will be utilized, and participants will have timed access to multiple LLMs through laptops provided by the organizers. They will earn points through a capture-the-flag-style point system that encourages testing a wide range of potential harms. At the end, the person with the most points will win a high-end Nvidia GPU.

The exercise aims to create a red-teaming process by which security experts can identify vulnerabilities or flaws in an organization's systems to improve overall security and resilience. Sven Cattell, the founder of AI Village, states that "The diverse issues with these models will not be resolved until more people know how to red team and assess them." 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.

AI researcher Simon Willison has previously written about the dangers of prompt injection, a technique that can derail a language model into performing actions not intended by its creator, making LLMs challenging to lock down. The DEF CON event aims to provide critical information to researchers and the public about the impacts of these models and to enable AI companies and developers to take steps to fix issues found in those models. The organizers plan to publish what they learn from the event to help others who want to try the same thing, creating a safer space for everyone.

DEF CON 31 will take place on August 10-13, 2023, at Caesar's Forum in Las Vegas.