Some of the 3D digital models created by researchers to depict lost neighborhoods in Columbus, Ohio, tell a clear story by placing the “ghosts” of houses that were demolished for freeway construction atop the roadways now occupying that land. But by talking to the people who lived in these communities, scientists are filling in historic gaps that technology can’t fill, adding trees, cars, people, photographs and life stories to the digitized infrastructure that was an initial focus of the work. The research team has published an update on the Ghost Neighborhoods of Columbus project in the International Journal of Digital Humanities, reporting on the technology workflow, including solutions to bottlenecks they’ve encountered, and acknowledging a slow start to community engagement that hums along nicely these days. The team aims to put both the digital work and story collection on a faster track.

Researchers have identified key components in large language models (LLMs) that play a critical role in ensuring these AI systems provide safe responses to user queries. The researchers used these insights to develop and demonstrate AI training techniques that improve LLM safety while minimizing the “alignment tax,” meaning the AI becomes safer without significantly affecting performance.