Atlas · GenAI 2026
Hallucination Detection
Hallucination detection & mitigation
conceptPeak: 2024Reliability & HallucinationAI consensus: 1/3
Prerequisites
RAG is the primary technique for mitigating hallucinations via grounding — understanding RAG informs detection strategies
Detecting hallucinations requires automated evaluation (faithfulness metrics, fact-checking) — eval frameworks are the detection tools
Recommended reference
Ji et al. (2023) 'Survey of Hallucination in Natural Language Generation' — ACM Computing Surveys; comprehensive taxonomy of hallucination types and mitigations
Notes from AI deep research
Anthropic Opus
Fact-checking, grounded generation, faithfulness. RAG = glowna mitygacja. Ji (2023) survey
Related skills
- → is part of: AI Output Verification(3/3)
- → is part of: LLM Evaluation Frameworks(3/3)
- → is subcategory of: AI Safety(2/3)
- ← is part of: AI Grounding & Citations(1/3)