🗂️ AI & Agents · View mindmap

Hallucination

A phenomenon where an AI system generates content that is not grounded in its training data or external knowledge sources, often appearing as fabricated or incorrect information.

Key Characteristics

• Non-factual output: Information presented as factual but unverifiable or false.
• Confidence mismatch: High confidence in incorrect statements.
• Contextual disconnect: Responses unrelated to input prompts.

Causes

• Data gaps: Insufficient or noisy training data.
• Model limitations: Overfitting or underfitting during training.
• Prompt engineering: Ambiguous or misleading user queries.

Mitigation Strategies

• Multi-agent systems: Combining multiple LLMs to cross-validate outputs.
• External verification: Integrating retrieval-augmented generation (RAG) for fact-checking.
• Human-in-the-loop: Manual review of critical outputs.

Related Concepts

• Retrieval-Augmented Generation
• prompt-engineering
• multi-agent-systems

References

• 2026 04 14 Anthropic multi agent deep Research agent

New Research

• Anthropic Multi-Agent Deep Research Agent (2026-04-14):
  • Designed to overcome hallucination via deep research agent flow.
  • Uses Flowise for structured workflows.
  • Addresses limitations of single LLM queries.
  • Tutorial by Leon van Zyl: https://www.youtube.com/watch?v=GPsKnsYJPiI.
  • GitHub repository: https://github.com/leonvanzyl/flowise-masterclass-2025/tree/master/Deep%20Research%20Agentflow.

Source Notes

• 2026-04-14: “But OpenClaw is expensive…”
• 2026-04-07: Next Evolution of Retrieval-Augmented Generation
• 2026-04-10: LiteParse - The Local Document Parser
• 2026-04-08: LiteParse: LlamaIndex
• 2026-04-22: Stanford