Hackers vs. AI at DEF CON
Once ChatGPT hit the market, hackers started to hack it and override its safeguards. OpenAI, Google, Microsoft, and others are working with the Biden administration to let hackers take a shot at testing the technology’s limits.
The companies want to know: How can chatbots be manipulated to cause harm? Will they share the private information we confide in them to other users? And why do they assume a doctor is a man and a nurse is a woman?
The companies want thousands of people for a mass hacking event planned for this year’s DEF CON hacker convention in Las Vegas. The companies hope the hackers will unearth problems that they can fix.
Some of the known problems include:
- ChatGPT, Microsoft’s Bing chatbot, or Google’s Bard have a tendency to fabricate info and present it as fact
- They emulate the cultural biases they’ve learned from being trained on what people have written online
- With the “grandma exploit,” users were able to get chatbots to tell them how to make a bomb by asking it to pretend it was a grandmother telling a bedtime story about how to make a bomb
This year’s event is the first to tackle the large language models that have attracted a surge of public interest and commercial investment since the release of ChatGPT. Companies that have agreed to provide models for testing include OpenAI, Google, Nvidia, and startups Anthropic, Hugging Face, and Stability AI. Building the platform for testing is another startup called Scale AI.
