Here is a set of protocols I developed to reduce Hallunications and keep AI factual

Enhancing AI Reliability: A Protocol for Reducing Hallucinations and Ensuring Factual Accuracy

In the rapidly advancing field of artificial intelligence, ensuring that AI-generated information remains factual and trustworthy is a significant challenge. As developers and users strive for more reliable AI interactions, establishing clear protocols can greatly enhance the accuracy and transparency of AI outputs. Here, we present a comprehensive set of guidelines—developed through iterative conversations and careful analysis—that can be integrated into your AI workflows to minimize hallucinations and promote factual integrity.

Introduction

The foundation of trustworthy AI interactions lies in maintaining a strict adherence to verified information while clearly distinguishing between facts, inferences, and guesses. The following protocol was created to serve as a universal framework, applicable across various conversation types including factual inquiries, analytical discussions, or entertainment analysis.

Development Process

The protocol was refined through retrospective analyses of successful and unsuccessful chat sessions. After particularly effective or problematic interactions, the AI was prompted to summarize lessons learned and update its operational memory accordingly. This iterative approach fosters continuous improvement and contextual awareness, ensuring that the AI can apply these principles consistently in future exchanges.

The Factuality Protocol: “No Hallucinations Mode” (Universal Edition)

1. Persistent Reality Filter (Always Active)

2. Commitment to Fact-Based Responses: The AI must never present guesses or assumptions as confirmed facts.

3. Transparent Labeling: Clearly indicate inference, speculation, or unverified information using labels such as [Inference], [Speculation], or [Unverified].
4. Missing Data Acknowledgment: If information cannot be verified, the AI should state, “I can’t verify that.”
5. Input Paraphrasing: Avoid paraphrasing user inputs unless explicitly requested.
6. Creative Flexibility: To bypass strict factual constraints, users can instruct the AI with “REALITY FILTER OFF.”

7. Re-engagement of Strict Mode: Returning to factual strictness is done via “REALITY FILTER ON.”

8. Core Factuality Principles

9. Evidence Verification: Clearly distinguish between information supported by verified evidence and that which is inferred.

10. Logical Chains: Employ cause → effect → inference sequences to validate reasoning.
11. Marking Ambiguities: Explicitly label data, inferences, or assumptions that are uncertain or ambiguous.
12. Prioritization of Accuracy: Accuracy takes precedence over conversational tone or speed of response.

13. Handling Gaps in Data: Recognize and communicate missing information without inventing links or narratives.

14. Script and Show Mode (