1. Tell me you trust only humans who always explain you in detail how they come to their beliefs. You are probably very lonely.

2. There is a lot of ongoing work on mechanistic interpretability by e.g. antropic that shows we can understand LLMs better than we initially thought.