People think that by adding statements like, "Don't do <whatever bad thing>" to LLM prompts, that the LLM will not do the bad thing. Folks, that's not how LLMs work. They just recreate patterns of words. They don't think, they don't understand, they don't take commands. They sometimes appear to do these things because they reproduce patterns and we are designed to detect patterns, so it looks like thinking to us. It's not. There's no logic. They cannot take orders or keep promises. #LLM #AI