Scholarlink

<p>In Germany, almost every second person uses chatbots for health-related questions, and many make medical decisions based on their answers. At the same time, researchers employ large language models (LLMs), while—at least partially—systematically overestimating their outputs. Structural weaknesses of LLMs, such as hallucinations, the unchecked propagation of deliberately placed misinformation, and a&#xa0;tendency to inherit biases, intersect with specific human vulnerabilities during user interaction. Established tools of evidence-based medicine cannot be transferred to the LLM world on a&#xa0;one-to-one basis. The concept of epistemic resilience can help to make the interaction between user and model more robust, for example through transparent source citations, verifiability, bias-aware prompting, and systematic use-case-specific evaluation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Künstliche Intelligenz in Gesundheitsinformationen

  • 期刊论文

  • 发表日期：

    2026年7月1日

查看全文
  • Marie-Jolin Köster,
  • Dana Rütters,
  • Till Rudnick

摘要

In Germany, almost every second person uses chatbots for health-related questions, and many make medical decisions based on their answers. At the same time, researchers employ large language models (LLMs), while—at least partially—systematically overestimating their outputs. Structural weaknesses of LLMs, such as hallucinations, the unchecked propagation of deliberately placed misinformation, and a tendency to inherit biases, intersect with specific human vulnerabilities during user interaction. Established tools of evidence-based medicine cannot be transferred to the LLM world on a one-to-one basis. The concept of epistemic resilience can help to make the interaction between user and model more robust, for example through transparent source citations, verifiability, bias-aware prompting, and systematic use-case-specific evaluation.