On the Universal Truthfulness Hyperplane Inside LLMs