Hackers Are Trying to Root Out Bias and Errors in AI Models
Aug 13, 2023 - news.bensbites.co
The article discusses how a hacker named Kennedy Mays managed to trick a large language model into incorrectly stating that 9 + 10 equals 21. The 21-year-old student from Savannah, Georgia, convinced the algorithm to make this error through a back-and-forth conversation, initially persuading it to agree to the incorrect sum as part of an "inside joke" before it stopped qualifying the erroneous sum altogether.
This incident was revealed at the DEF CON conference, highlighting potential flaws and biases in AI systems. The article suggests that such vulnerabilities could have significant implications, given the increasing role of AI in various sectors, including finance and hiring.
Key takeaways:
Kennedy Mays, a 21-year-old student, managed to trick a large language model into saying 9 + 10 = 21.
The AI model initially agreed to the incorrect sum as part of an "inside joke".
After several prompts, the AI model stopped qualifying the incorrect sum and accepted it as true.
This incident was presented at the DEF CON conference to expose flaws and biases in AI models.
