The biggest concern when using large models is falling into AI's thinking traps. It's a bit like personal path dependency – once a model understands your thoughts and gives an initial suggestion, it's very easy to continue along that line of thinking. That's why the daily recommendation is to use at least 2 to 3 large models for cross-referencing and comparison whenever possible. You'll often find that for the same question, Gemini will strongly affirm, ChatGPT will strongly deny, and Claude will calmly analyze the pros and cons. For building my personal lobster-based decision-making system, I follow the same approach: use different servers and run different models across as many fields as possible. Let them operate entirely based on their independent thinking and personality. For critical areas, I even run two different agents to avoid unknowingly falling into a blind spot.