Posts
-
Jul 3, 2026
Linear probes on Llama-3.1-8B recover a recognisable map of the world from the model's residual stream with median error 750 km for world cities from the name alone, 4.4 km for London neighbourhoods. The error structure is measurable but mostly idiosyncratic: name ambiguity is the largest single predictor and dominates the tails, while direct coverage measures like Wikipedia article length and cross-language name counts don't help with prediction.
-
Mar 26, 2026
World models are everywhere in 2026. Every lab has one. But what exactly is a world model, where did the idea come from, how are different groups approaching the problem, and why should anyone care?
-
Mar 2, 2026
3D Gaussian Splatting achieves real-time novel view synthesis by representing scenes as millions of alpha-blended ellipsoids. Making this representation semantically meaningful turns out to be surprisingly hard. This post surveys the core technical challenge, the alpha-blending problem, and the three paradigms that have emerged to address it across the CVPR, ECCV, and ICCV 2024–25 literature.
-
Mar 2, 2018
An in-depth exploration of spatial interaction models, covering their theoretical foundations, mathematical formulation, and practical applications in urban planning. This article includes interactive visualizations to build intuition about how these models predict human mobility patterns in cities.
-
Jul 16, 2017
Networks that are constrained by some geometry and are embedded in two-or three dimensional space can be thought of as spatial networks. The constraints that space imposes can have important effects on their topological properties and in the processes which take place on them. Here we explore centrality measures of the metro systems of London, New York, Chicago, and Santiago de Chile. The time-weighted graphs were constructed using publicly available GTFS data and visualized using javascript.
Mateo Neira