Gurnee & Tegmark (2024)

Language Models Represent Space and Time

arXiv

Find It

DOI

Abstract

The capabilities of large language models (LLMs) have sparked debate over whether such systems just learn an enormous collection of superficial statistics or a set of more coherent and grounded representations that reflect the real world. We find evidence for the latter by analyzing the learned representations of three spatial datasets (world, US, NYC places) and three temporal datasets (historical figures, artworks, news headlines) in the Llama-2 family of models. We discover that LLMs learn linear representations of space and time across multiple scales. These representations are robust to prompting variations and unified across different entity types (e.g. cities and landmarks). In addition, we identify individual "space neurons" and "time neurons" that reliably encode spatial and temporal coordinates. While further investigation is needed, our results suggest modern LLMs learn rich spatiotemporal representations of the real world and possess basic ingredients of a world model.

Reference

Gurnee, W., & Tegmark, M. (2024). Language Models Represent Space and Time. arXiv preprint arXiv:2310.02207. https://doi.org/10.48550/arxiv.2310.02207

What Rests on This

1 mechanism cites this source. Each finding below is quoted as it appears on that mechanism's own page.

Temporal Placement Memory Processes

Measured in AI

For what a model learned in training, the dates of historical figures' deaths, works of entertainment and news headlines could be read out of Llama 2's internal activity with R² between 0.56 and 0.89, rising with model size. No human dating data were collected on the same items, so there is no common scale with human event dating and its telescoping errors