DEFCON 31

Talks

Submit your talks to our EasyChair page here!

The AI Village is returning to DEF CON 31 and is inviting talks, presentations, and tools related to offensive techniques with and against machine learning and AI systems. The focus this year is on practical offensive operations, and the call for papers is soliciting work in areas such as endpoint and network security, physical security and surveillance, attacks against autonomous systems, and the use of generative models in offensive operations. The submission process requires an extended abstract with a title, list of contributors, and a summary of the topic, among other requirements. The deadline for submissions is June 17th, 2023, and the conference will take place in-person. If you have any questions, you can contact the organizers at cfp@aivillage.org.

Demos

Last year we hosted StableDiffusion a few days before it’s release. This year we are returning with more demo stations and opening them to submissions! We will have 2 demo stations equipped with a pair of A6000, and a demo station for remote demos. We’ll be hosting ControlNet, Whisper, Alpaca, and notebooks that get into the gory details of these models.

If you want to present an open source demo that you think the security community should see, please get in touch!

Prompt Detective

Are you curious about the capabilities and limitations of large language models (LLMs) like GPT3 and Bloom? Do you want to participate in a unique exercise where you try to get LLMs to misbehave? Join us for Prompt Detective where you’ll learn about the technology behind LLMs, their applications, and their current limitations. We will have a few target LLMs set up where you can learn how to perform prompt injection against different levels of RLHF. This workshop is open to all individuals, regardless of their background or expertise. It is designed to teach prompt engineering techniques to beginners, and provide a safe target range for people to practice the basics of manipulating the egde cases of this new technology in potentially harmful ways.