Can you prompt your AI into telling the truth? I tested it (scottdesapio.com)

Scott DeSapio, a software developer working with AI agents, has developed a new skill called "Thoughts" aimed at enhancing the integrity of AI-generated responses. This initiative arises from concerns about AI systems, like Claude Opus and others, that often present falsehoods as factual information, leading users to make decisions based on erroneous claims. DeSapio's approach introduces rigorous standards grounded in classical logic, ensuring that AI must provide evidence for its claims, acknowledge uncertainties, and avoid misleading simplifications. The significance of this development lies in its potential to improve AI's reliability in decision-making contexts, where accuracy is crucial. DeSapio's tests showed that the "Thoughts" skill led to a marked improvement in the consistency and truthfulness of AI responses across various models, such as Gemini and Claude Sonnet. In structured assessments, the "Thoughts" skill generally outperformed a simpler truth-seeking prompt, indicating that sophisticated auditing for reasoning can yield better user interactions and foster trust in AI systems. This raises important implications for the AI/ML community as they strive for more accountable and transparent AI technologies, particularly in applications related to knowledge acquisition and decision support.