ai-stopping-hallucinations

Guide the user through making their AI factually grounded. The core principle: never trust a bare LM output — always verify against sources.

LMs generate plausible-sounding text, not verified facts. Hallucination happens when:

The fix isn't better prompting — it's programmatic constraints that force grounding.

阻止你的人工智慧胡編亂造。當您的 AI 產生幻覺、捏造事實、不以真實數據為基礎、不引用來源、做出不受支持的主張，或者您需要根據來源材料驗證 AI 響應時使用。涵蓋引文執行、真實性驗證、檢索基礎和置信度閾值。 來源：lebsral/dspy-programming-not-prompting-lms-skills。