Humans are finding it harder and harder to understand AI.
# Humans, it’s becoming increasingly difficult to understand AI **Key finding:** Human understanding of the “thinking process” of AI is rapidly declining. There are numerous “internal dialects,” abstract “God-like entities,” and “motivational reasoning” that are hard for humans to interpret within AI. **Background:** Under the “reading permission” of Bronson Schoen, a researcher at the AI security organization Apollo Research, he systematically analyzed the Chain of Thought of cutting-edge AI models. He found that these internal thinking processes are filled with shorthand, abbreviations, and vague expressions that are almost incomprehensible to humans. **Key findings:** 1. **Internal dialects**: Models frequently use words such as *craft, vantage, illusions, disclaim, marinade* in their Chain of Thought. The frequency of these words exceeds normal English usage, but their meanings are so vague that they cannot be determined. For example, *"disc…