From Comics to Code: Building MyZubster One Piece of Evidence at a Time

What happened

The article describes the development of MyZubster, an AI system that focuses on evidence-based verification. It outlines the challenges faced during development, such as ensuring that AI does not generate incorrect answers based on incomplete or misleading evidence. The system uses a combination of semantic retrieval, local AI inference, and a RAG layer to connect questions with retrieved sources. The project also emphasizes the importance of separating information retrieval from truth verification and the role of decentralization in ensuring verifiable evidence.

Why it matters

For AI developers and technology professionals, this article highlights the importance of building systems that prioritize evidence over speculation. It provides insights into the challenges of ensuring AI accuracy and the role of decentralized systems in verifying data. This is particularly relevant for developers working on AI applications that require high levels of trust and transparency.

• AI systems should prioritize evidence over speculation to ensure accuracy.
• Separating information retrieval from truth verification is essential for building reliable systems.
• Decentralization requires a solid foundation of verifiable evidence and provenance.
• Local AI inference can be useful when interpretation and synthesis are required, but should not be used as an intermediary for every piece of stored evidence.
• The system should return authoritative descriptions verbatim when the evidence is available.

The bigger picture

The practical impact of MyZubster is that it enables users to verify the origin and accuracy of information, reducing the risk of AI-generated misinformation. It also sets a precedent for building systems that prioritize evidence over speculation, which is crucial for applications requiring high levels of trust and transparency.

The article concludes that MyZubster is a project that aims to build a system capable of distinguishing between what was observed, recorded, retrieved, verified, and generated. It emphasizes the importance of evidence-based verification and the role of decentralization in ensuring trust and transparency in AI systems.

Technical details

• Qdrant is used for semantic retrieval of evidence.
• Ollama provides local AI inference capabilities.
• A RAG (Retrieval-Augmented Generation) layer connects questions, retrieved sources, and generation.
• The system prioritizes exact observation IDs over semantic similarity.
• An observation-index recovery mechanism is used to rebuild the search index from persisted observations.
• The system uses an OpenAI-compatible API to expose the RAG system.

FAQ