Please tell me more. When I ask an LLM a question, and get a text response, can that response incorporate non-textual information from visual training data?