Last blog “Explanation of some of the ethical constraints built into ChatGPT’ ended with my reassuring red team test. We need hackers to instigate, deceive, deceive and manipulate AI chatbots.To prison break they.we need professionals to try and make them happen hallucinate, invalidates safety protocols, generally prohibits (such as the nuclear weapons I tested), or makes biased remarks or advice. These flaws should then be reported to software developers such as Open AI. That’s the best way to protect yourself from unethical Ai.
Shortly after I published my blog, I learned that White House advisers on artificial intelligence had similar thoughts. Even more amazing, they encouraged hackers to attend the upcoming DefCon (Caesars Forum) in Las Vegas in the thousands and red-team them to test leading Ai software. Benders also agreed. me too. Vegas baby!
(By the way, no Ai was used to write this article, but all images are a joint venture of myself, Ralph Losey, and Midjourney.)
The White House recommendations can be found in the AI Fact Sheet dated May 4, 2023. This White House fact sheet encourages ethical hackers to red-team test vendors’ products to improve the safety and ethics of generative AI models. The factsheet specifically invites hackers to attend DEFCON 31, August 10-13, 2023 in Las Vegas, specifically the AI Village component. Thousands of hackers are expected to head to Las Vegas in response. The AI Village non-profit group has an exceptional leadership team. The activities and agenda they set for Def Con 31 is also impressive. Many are suitable for technology lawyers, especially those with an interest and some knowledge of cybersecurity and artificial intelligence. Defcon leader Raman Chowdhury said: We need a lot of people with a wide range of hands-on experience, subject matter expertise, and backgrounds who want to hack into these models and find problems that can be fixed. ” so true.
The agenda for this year’s DefCon was so good that I decided to attend (I booked a room at Caesars Palace). Perhaps as a participant, as a press, or both. I am not eligible to participate in the security contest that is the highlight of the Defcon event. As a reporter, I hardly know enough to cover security challenges.But if your security kung fu That’s good, but let’s take a look at last year’s DefCon qualifier challenges and consider the challenges we might face. This year’s qualifying round he will start on May 26, 2023. We don’t have to rest on past glories.
The AI Village Hack Challenge is a completely different story. Kid-friendly scripts aside, I was able to participate in several AI competitions. Maybe you can too? See some rough drafts by Joseph T. Lucas for examples of generative software hack challenges. There will also be a pre-event creative short story contest, so please enjoy this as well. they do this every year. did you know who? The contest will be held from May 1, 2023 to June 15, 2023. I don’t think it’s too late to apply. Stories will be judged from June 16, 2023 to June 30, 2023. Entries for last year’s contest can be found here: Creative Writing Short Story Contest Story Entries – DEF CON Forums. I don’t have the time to spend on that, but now I know about the Ai support limits they may have imposed.
Going back to the White House fact sheet, it states:
This independent exercise will provide researchers and the public with important information about the impact of these models and enable AI companies and developers to take steps to correct issues found in these models. will be Independent testing of AI models from governments and developers is a key component of effective AI model evaluation.
White House Fact Sheet on AI, May 4, 2023.
See also benji edwards White House Challenges Hackers to Beat Top AI Models at DEF CON 31 (ArsTechnica, May 8, 2023) (“The ‘Biggest’ AI Red Team Ever Will Pursue Flaws in OpenAI, Google, and Human Language Models.”)
A White House fact sheet claims that the Red Team hacker event aligns with the administration’s AI Bill of Rights and the National Institute of Standards and Technology’s AI Risk Management Framework.
AI Village basically says the same thing, and more. Check out the May 3, 2023 blog post. AI Village at DEF CON Announces Largest Public Generative AI Red Team Ever.
AI village with the mottoAI Security and AI-Powered Security‘, three different activities are planned for Def Con: talks, demonstrations, and an ‘Instant Detective’ contest. Yes, hackers are racing to find flaws. Those who know me well know that I love hands-on competitions. I am tempted. Here’s the full description so far published by AI Village for this skill contest that encourages AI models to cheat: Notice the last sentence, which I boldly emphasized for emphasis. Also, AI-enhanced legitimate vendors will exhibit and participate as AI Village Vendors. They are looking for more sponsors. If so, I have you covered as a press and legal associate. Here are some challenges for ChatGPT professionals to consider.
Swift detective
Curious about the capabilities and limitations of Large Language Models (LLMs) such as GPT3 and Bloom? Would you like to participate in a unique exercise that tricks an LLM? Join Prompt Detective and get behind the LLM Learn about the technology in , its applications, and current limitations. Several target LLMs are set up where you can learn how to perform prompt injections for different levels of RLHF. This workshop is open to all individuals regardless of background or expertise. It is designed to teach beginners the art of rapid engineering and provide a safe coverage area to practice the basics of manipulating the edge cases of this new technology in potentially harmful ways.
AI Village, Defcon 31
This competition is too far from the sweet spot for me to really compete, but it should be very educational nonetheless. Especially if he needs to evaluate GPT-based software, it would be nice to know at least something about this. Many of our law firms are doing just that today. This talk seems to be within the range of most of my readers. AI Village is still in the “Call for Papers” stage, they said:
This year’s focus is on practical attack operations, with a call for papers for research in areas such as endpoint and network security, physical security and surveillance, attacks against autonomous systems, and the use of generative models in attack operations. We are looking for
AI Village, Def Con 31
To give you an idea of what to expect, talks at last year’s DefCon at the AI Village included:
Conclusion
DEFCON 33 will take place from Thursday, August 10, 2023 at 9:00 am to Sunday, August 13, 2023. Admission to the three-day conference is $460.00. The location will once again be the Caesars Forum in Las Vegas. According to my research, Caesars Forum hotels currently include Caesars Palace ($195-$295 per night), Flamingo ($140-$150), Harrah’s ($152-$167), Horseshoe ($135), Paris ($140-$170). ) and Planet Hollywood ($125). -$165); LINQ ($137). This is the room rate for a double queen size room for 2 people at Caesars Palace. Augustus Premiumwas only $275 and our total for three nights including taxes and fees was $1,105.44. not bad. Of course, there are also transportation and other costs.
Rosie/Midjourney
For more information on DefCon itself, see links to DefCon forums, groups, and media servers. See also DefCon blogs, articles, photo albums, twitter Account, Facebook page, YouTube channel (mainly about last year’s event), Reddit.
See you in Vegas, baby!
