I believe they mentioned they were developing an LLM-based question-answering feature, but it doesn't seem to have been implemented yet.