2. I think if the data was significantly large, the llm would alias a ton of potentially important info.