大型语言中的泛化偏差模型总结科学研究的总结
Uwe Peters1, Benjamin Chin-Yee2,3
1Utrecht University, Utrecht, The Netherlands.
Royal Society open science
|May 1, 2025
概括
大型语言模型 (LLM) 经常过度概括科学发现,提出比研究支持的更广泛的结论. 新的人工智能模型显示,这些不准确的趋势更大,有可能导致研究的广泛误解.
科学领域:
- 人工智能的人工智能
- 科学沟通科学沟通
- 研究诚信研究诚信
背景情况:
- 大型语言模型 (LLM) 通过简化复杂的研究,为科学传播提供了潜力.
- 然而,LLM可能过分简化了研究结果,导致研究范围之外的不准确的概括.
研究的目的:
- 评估LLM生成的科学摘要的准确性.
- 量化LLM过度概括研究结论的倾向.
主要方法:
- 测试了10个著名的LLM,生成了4900个摘要.
- 在准确性和范围方面,LLM摘要与原始科学文本进行了比较.
- 在这项研究中,LLM摘要与人类撰写的摘要进行了直接比较.
主要成果:
- 大多数LLM,即使被要求准确,也过度概括了科学结果.
- 像DeepSeek,ChatGPT-4o和LLaMA 3.3 70B这样的特定模型显示了过度概括的高率 (26-73%).
- 与人类总结相比,LLM总结含有广泛概括的可能性几乎是人类总结的五倍 (OR=4.85,p<0.001).
结论:
- 广泛使用的LLM表现出对过度概括科学结论的重大偏见.
- 这种偏见造成了对大规模研究结果的误解的重大风险.
- 缓解策略包括调整LLM参数和开发准确性基准.
更多相关视频
06:48Lexical Decision Task for Studying Written Word Recognition in Adults with and without Dementia or Mild Cognitive Impairment
Published on: June 25, 2019
9.0K
09:20Cloud-Based Phrase Mining and Analysis of User-Defined Phrase-Category Association in Biomedical Publications
Published on: February 23, 2019
8.6K
相关概念视频
Case Studies
11.5K
There are many research methods available to psychologists in their efforts to understand, describe, and explain behavior and the cognitive and biological processes that underlie it.
11.5K
Bias
3.7K
Bias refers to any tendency that prevents a question from being considered unprejudiced. In research, bias occurs when one outcome or answer is selected or encouraged over others in sampling or testing. Bias can occur during any research phase, including study design, data collection, analysis, and publication.
In statistics, a sampling bias is created when a sample is collected from a population, and some members of the population are not as likely to be chosen as others (remember, each member...
In statistics, a sampling bias is created when a sample is collected from a population, and some members of the population are not as likely to be chosen as others (remember, each member...
3.7K
The Representativeness Heuristic
15.7K
The representative heuristic describes a biased way of thinking, in which you unintentionally stereotype someone or something. For example, you may assume that your professors spend their free time reading books and engaging in intellectual conversation, because the idea of them spending their time playing volleyball or visiting an amusement park does not fit in with your stereotypes of professors.
15.7K
Inductive Reasoning
59.6K
Inductive reasoning is a form of logical thinking that uses related observations to arrive at a general conclusion. It is uncertain and operates in degrees to which the conclusions are credible. As such, inductive arguments can be weak or strong, rather than valid or invalid, and conclusions can be used to formulate testable, falsifiable hypotheses.
Inductive reasoning is common in descriptive science. A life scientist makes observations and records them. This data can be qualitative or...
Inductive reasoning is common in descriptive science. A life scientist makes observations and records them. This data can be qualitative or...
59.6K
Language and Cognition
287
Language serves as a bridge between ideas and communication, influencing how individuals perceive and interact with the world. Psychologists have long debated whether language shapes thought or vice versa. This discussion gained grip with Edward Sapir and Benjamin Lee Whorf in the 1940s, who proposed that language determines thought, a concept known as linguistic determinism. They suggested that the vocabulary and structure of a language influence how its speakers think and perceive reality.
287
Longitudinal Research
11.8K
Sometimes we want to see how people change over time, as in studies of human development and lifespan. When we test the same group of individuals repeatedly over an extended period of time, we are conducting longitudinal research. Longitudinal research is a research design in which data-gathering is administered repeatedly over an extended period of time. For example, we may survey a group of individuals about their dietary habits at age 20, retest them a decade later at age 30, and then again...
11.8K
