On the Role of Citations in Preference Data
Researchers studied how humans and language models evaluate citations in scientific question answering tasks. They found that humans prefer diverse but limited citations, while language models show varying preferences depending on the data and model. The study's findings have implications for collecting preference data.
Researchers studied how humans and language models evaluate citations in scientific question answering tasks. They found that humans prefer diverse but limited citations, while language models show varying preferences depending on the data and model. The study's findings have implications for collecting preference data.
---
Why it matters: Understanding how to effectively use citations in AI systems is crucial because it can impact the credibility of model outputs and prevent 'hallucinations' where models make up information without a basis. This research helps advance the development of more reliable AI systems.
Source: https://arxiv.org/abs/2608.21376
This article was originally published at: https://arxiv.org/abs/2608.21376