LLM complexity depends on its use case - if it’s a chat window, this kind of dichotomy wouldnt work (Unless I am missing something in this thread) The model wi

LLM complexity depends on its use case - if it’s a chat window, this kind of dichotomy wouldnt work (Unless I am missing something in this thread) The model will be treating all text as input, whether it’s tagged as instruct or data. You could find a way to abuse either entry point.