DistRAG: Towards Distance-Based Spatial Reasoning in LLMs

Nicole R. Schneider, Nandini Ramachandran, Kent O’Sullivan, Hanan Samet · 2025

Read DistRAG: Towards Distance-Based Spatial Reasoning in LLMs by Nicole R. Schneider, Nandini Ramachandran, Kent O’Sullivan, Hanan Samet (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-03424.txt.

Description

Nicole Schneider and three co-authors give language models a way to reason about distance in a 2025 paper. Models are weak at judging how far apart places are, which hurts tasks like trip planning and local search tips. Their method, DistRAG, stores real distances between towns in a graph.

When a question arrives, the system pulls the part of the graph that matters and hands it to the model as context. The model then answers with true distance facts instead of guessing from text patterns alone. The authors call this a flexible first step toward a basic world model for models. It can cover any set of places the graph holds. Tests show the model answering distance questions it could not handle before.

Catalog record

CREATORNicole R. Schneider, Nandini Ramachandran, Kent O’Sullivan, Hanan Samet
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE19 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-03424

Browse the full library · About the Tower of Babel