Authors claim they just figured out how to predict LLM hallucinations (github.com)