The thing that strikes me about LLMs is that they have been created to chat. To converse. They’re partly influenced by Turing tests where the objective is to convince someone you’re human by keeping up a conversation. They weren’t designed to create meaningful content or factual content.
People still seem to want to use chat GPT to create something, and fix the accuracy as a second step. I say go back to the drawing board and create a tool that analyses statements and tries to create information based on trusted linked open data sources.