Tasks that require actual logic or concepts rather than retrieving text seem to defeat LLMs. Rather like me trying to get an LLM to write the svg code for an INVERTED pink triangle. It couldn't invert it. It cannot understand concepts like humans can.