LLMs are rubber ducks that can talk back to you

My experience of the last two years is that this works every bit as much for conceptual or methodological problems as it does for programming ones: