My problem with the approaches to understanding other than reading the actual code is LLMs hallucinate what code actually does based on what the intent of the code is