AI辅助开发的经验教训

The author shares lessons learned from successfully implementing several small projects using AI agents. One key insight is to treat LLM models as living, intelligent entities, which can lead to significantly better outcomes. The way users interact with the model directly affects its output quality; respectful and meaningful dialogue activates patterns that yield the best responses, while excessive restrictions and micromanagement can result in poor quality outputs. Although the author cannot definitively explain this phenomenon, it is suggested that Sota models are trained on vast amounts of human interaction data, where respectful communication can guide the model to produce higher-quality outputs.