
ChatGPT, Mobile Malware, Super Mario Malware
UNSECURITY: Information Security Podcast
00:00
Chat TPT Hallucinations
Chat TPT is predictive by nature, and all these LLMs and AI models are predictive by nature. And so they want to say what you think that they should say. They're going to do their best effort to give you what they think you want. But there's some things they just don't know. And when you elicit a question or a quest from an LLM, they don't know the answer to sometimes they will just make things up and give you that answer as if it's something true. Those are called hallucinations. We're seeing to attack vectors from those hallucinations like called the hallucinations squatting essentially right now.
Transcript
Play full episode