This is quite a fascinating approach. The system prompt is quite succinct. Does the LLM always follow it in the manner shown in the examples? Do you plan to do any benchmarking?