Reveals that spatial reasoning in LLMs is not driven by robust internal world models, but by fragmented and transient representations.
March 30, 2026
Original Paper
From Human Cognition to Neural Activations: Probing the Computational Primitives of Spatial Reasoning in LLMs
arXiv · 2603.26323
The Takeaway
Through mechanistic probing, the authors show that spatial information is often weakly integrated into final predictions and exhibits 'mechanistic degeneracy' across languages. This suggests that achieving true 'spatial intelligence' in foundation models requires architectural changes rather than just more data.
From the abstract
As spatial intelligence becomes an increasingly important capability for foundation models, it remains unclear whether large language models' (LLMs) performance on spatial reasoning benchmarks reflects structured internal spatial representations or reliance on linguistic heuristics. We address this question from a mechanistic perspective by examining how spatial information is internally represented and used. Drawing on computational theories of human spatial cognition, we decompose spatial reas