| | Meet the speakers for the 2026 Semantic Layer Summit. Enterprise leaders from Papa Johns, Carrefour, Blue Yonder, Snowflake & more share what it takes to make AI work at enterprise scale. Register free today!
*Message from this week's sponsor, AtScale |
| | |
| | 🔗 Note: if you don’t see links in this email, it’s also available on the web: dataelixir.com/archive/issue-567.html |
| | |
| | LLMs can’t do math. They’re language models. Percepta’s answer is to stop predicting tokens and start executing programs inside the transformer. Their proof of concept compiles a full VM into the model weights. It’s wild, it’s limited, and it’s one of the more interesting ideas in recent transformer research.
Percepta | Christos Tzamos |
| | |
| | Jacobi fields describe how nearby paths on a curved space drift apart, linking small changes at the start to how far apart you end up. This post makes that connection visual and shows why it matters for ML models that operate on manifolds.
Olga Zaghen |
| | |
| | Merges look simple until they aren’t. This guide walks through joins step by step, using SQL and Stata to show how duplicate keys and many-to-many merges quietly break analyses and how to catch them early.
Florian Oswald |
| | |
| | This post reads more like a story than a tutorial. It starts with the idea of “character” and follows that step by step until it becomes measurable, using data to uncover a socioeconomic divide hiding in plain sight.
Lauren Leek |
| | |
| | | | ggauto encodes visualization best practices directly into ggplot2, automatically selecting chart types and styling based on your data. The result is fewer questionable charts and more consistent, readable outputs by default.
Nicola Rennie |
| | |
| | This tutorial takes a physics-heavy idea, strips it down to something tractable, and implements it end-to-end in R using the rayverse stack. The realism is impressive, but what stands out is the workflow. You see how to go from physical model to rendered 3d scene, and then refine it through visual iteration.
Tyler Morgan-Wall |
| | |
| | | Working with tables on the web usually means copy, paste, clean, repeat. This extension skips all that. It overlays visual signals directly on the data so you can scan patterns, spot outliers, and explore without ever leaving the page.
Chrome Web Store | Ranjith Kumar Saila |
| | |
| | Cities are full of graphs, but most geospatial tooling doesn’t treat them that way. City2Graph converts streets, transit, OD flows, and spatial relationships into graph structures you can analyze with NetworkX or push into PyTorch Geometric. It's a solid bridge between GIS workflows and graph ML.
c2g-dev |
| | |
| | | Some ideas don’t exist until you can write them down. This article explores how mathematical notation evolved alongside what people could actually compute, and how better representations open up entirely new ways of thinking.
Quanta Magazine | John Pavlus |
| | |
| | | | This week's issue sponsored by: | | |
|
