the biggest problem with LLMs is that they're trained on massive amounts of existing data, but they don't actually understand the context or implications of what they're spitting out, they're just really good at pattern matching.