5 accepted papers
Current AI safety frameworks, which often treat harmfulness as binary, lack the flexibility to handle borderline cases where humans meaningfully disagree. To build more pluralistic systems, it is essential to move beyond consensus and instead understand where and why disagreements arise. We introduc…
Preference alignment via reward models helps build safe, helpful, and reliable large language models (LLMs). However, subjectivity in preference judgments and the lack of representative sampling in preference data collection can introduce new biases, hindering reward models’ fairness and equity. In…
Empathy serves as a cornerstone in enabling prosocial behaviors, and can be evoked through sharing of personal experiences in stories. While empathy is influenced by narrative content, intuitively, people respond to the way a story is told as well, through narrative style. Yet the relationship betwe…
Most NLP work on narrative detection has focused on prescriptive definitions of stories crafted by researchers, leaving open the questions: how do crowd workers perceive texts to be a story, and why? We investigate this by building StoryPerceptions, a dataset of 2,496 perceptions of storytelling in…
Story detection in online communities is a challenging task as stories are scattered across communities and interwoven with non-storytelling spans within a single text. We address this challenge by building and releasing the StorySeeker toolkit, including a richly annotated dataset of 502 Reddit pos…