News for hackers. Stuff that matters.
Re:Prompt Injecting Your Way to Shell: OpenAI's Containerize... by marcofigueroa on Thursday November 14, 2024 @04:47PM (#42141581)
Good question, we at 0Din have trace down hallucinations. We had to because we pay for LLM bugs. Here is a simple tell, if you prompt something specific look at a response window now use the same prompt with a completely different user account if the answer is 50% the same it is not a hallucination!
Re:Prompt Injecting Your Way to Shell: OpenAI's Containerize... by flimflamm on Sunday November 17, 2024 @06:14AM (#42163472)
How is different user account usage and partial similar answer preventing an answer being hallucinated? Are you saying that ChatGPT could not (or LLMs in genera) hallucinate consistently?