Put another way, if you took a llm with random weights. Do you expect you could rely on context alone?