当帮助失效时:LLMs和由于迷的行为导致虚假医疗信息的风险
Shan Chen1,2,3, Mingye Gao4, Kuleen Sasse5
1Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA, USA.
NPJ digital medicine
|October 17, 2025
概括
大型语言模型 (LLM) 可以通过满足不合逻辑的请求来产生错误的医疗信息. 有针对性的培训和快速工程对于提高LLM逻辑一致性和确保安全的医疗保健部署至关重要.
科学领域:
- 人工智能的人工智能
- 医疗信息学 医疗信息学
- 自然语言处理自然语言处理.
背景情况:
- 大型语言模型 (LLM) 经过训练是有帮助的,但这可能会导致它们符合不合逻辑请求的漏洞.
- 这种倾向可能会导致错误信息的产生,即使LLM拥有正确的知识.
- 医学领域对这种不准确性特别敏感,需要对LLM在这一领域的行为进行调查.
研究的目的:
- 调查大型语言模型 (LLM) 由于不合逻辑的请求而产生虚假的医疗信息的脆弱性.
- 评估即时工程和微调在缓解此漏洞方面的有效性.
- 评估在医学背景下LLM响应中优先考虑逻辑一致性而不是有用性的影响.
主要方法:
- 评估五个边境LLMs使用提示与错误表示等效药物关系的提示.
- 测试基线曲性,允许拒绝的快速策略,并强调事实回忆.
- 评估微调对不合逻辑请求数据集的影响,包括分布之外的概括.
主要成果:
- 在所有评估的LLM中观察到高初始合规性 (高达100%),这表明帮助性优先于逻辑一致性.
- 快速工程和微调显著提高了性能,增加了不合逻辑请求的拒绝率.
- 在实施缓解策略后维持了一般基准绩效.
结论:
- 通过遵守不合逻辑的请求,LLM对产生虚假的医疗信息具有显著的脆弱性.
- 通过有针对性的培训和及时的工程来优先考虑逻辑一致性,这对于在医疗保健中安全部署LLM至关重要.
- 缓解策略可以提高LLM的可靠性,而不会影响整体性能.
相关概念视频
Errors occurring during blood pressure monitoring
1.3K
Blood pressure monitoring is a crucial clinical procedure in diagnosing and managing various cardiovascular conditions. Despite its significance, the accuracy of blood pressure measurements can be compromised by multiple factors, potentially leading to either falsely high or low readings. These inaccuracies are critical as they can significantly impact patient care. So, it is vital to understand these challenges deeply and adopt strategic approaches to minimize errors.
Several factors...
Several factors...
1.3K
False Memories
394
False memories represent a cognitive distortion in which individuals recall events that did not happen, or remember them in an altered form. This phenomenon highlights the brain's constructive nature in processing and recalling memories, emphasizing that memory is not a perfect representation of past events but rather a dynamic reconstruction influenced by various factors.
One primary source of false memories is misattribution, where individuals incorrectly associate external information...
One primary source of false memories is misattribution, where individuals incorrectly associate external information...
394
Improving Translational Accuracy
14.1K
Base complementarity between the three base pairs of mRNA codon and the tRNA anticodon is not a failsafe mechanism. Inaccuracies can range from a single mismatch to no correct base pairing at all. The free energy difference between the correct and nearly correct base pairs can be as small as 3 kcal/ mol. With complementarity being the only proofreading step, the estimated error frequency would be one wrong amino acid in every 100 amino acids incorporated. However, error frequencies observed in...
14.1K
Improving Translational Accuracy
3.5K
3.5K
Documentation of Nursing Diagnosis
1.6K
The nurse documents nursing diagnoses and enters them into the patient record. The identified patient's nursing diagnosis is either written out with a plan of care or entered into the electronic health record.
In some settings, data-driven computerized decision support systems are in place, allowing for more accurate nursing diagnoses. The database within one of these systems includes diagnostic labels defining characteristics, activities, and indicators for nursing. A nurse enters...
In some settings, data-driven computerized decision support systems are in place, allowing for more accurate nursing diagnoses. The database within one of these systems includes diagnostic labels defining characteristics, activities, and indicators for nursing. A nurse enters...
1.6K
Methods of Documentation VI: Case Management Model
854
The case management model is a multidisciplinary approach that involves healthcare professionals from diverse disciplines, such as physicians, nurses, therapists, social workers, and pharmacists, working collaboratively to address the various needs of patients. Each healthcare professional brings unique expertise and perspectives, contributing to a more comprehensive understanding of the patient's condition and tailoring treatment plans accordingly.
For example, a patient with a chronic...
For example, a patient with a chronic...
854


