| | AI systems are only as reliable as the context behind them. This free summit looks at how teams at Carrefour, Blue Yonder, Anthropic, and others are using semantic layers to keep metrics aligned, ground AI in shared definitions, and avoid drift across systems.
*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-572.html |
| | |
| | This is a nice write-up on turning AI from a helpful text box into something closer to a working collaborator. The examples are mostly Claude Code, but the core lessons are broader: context matters, taste matters, and good systems make verification easier.
Eugene Yan |
| | |
| | Prediction markets are supposed to aggregate useful information, but most of the action these days is in sports, crypto, elections, and entertainment. This article takes a look at Kalshi and Polymarket data to find where markets actually help decision-making, where they mostly duplicate existing signals, and why AI forecasters may change the story.
Asterisk Magazine | Dan Schwarz |
| | |
| | Steering a model isn’t always as simple as moving in a straight line. This post argues that concepts inside neural networks often form curved structures, and understanding those shapes may be key to interpreting, debugging, and steering model behavior.
Goodfire | Atticus Geiger, et al. |
| | |
| | This post walks through how a Finnish-English dictionary went from a 3 GB database to a tiny static binary, and why matching the data structure to the query pattern can matter more than using a familiar database.
Andrew Quinn |
| | |
| | dbt usually lives in the analytics engineering world, while R’s {targets} is more at home in reproducible data science. This post rebuilds the same pipeline in both, showing how tool choice changes the shape of a project: SQL-first and warehouse-oriented, or R-native and analysis-driven.
Jonathan Carroll |
| | |
| | | | Not every data science career runs through FAANG, frontier models, or PhD-level methods. This Reddit thread is a useful reminder that a lot of durable career value comes from translating messy business questions, choosing simple methods, communicating clearly, and making decisions easier.
Reddit / datascience |
| | |
| | | Most forecast charts make the future look cleaner than it is. This post argues for visualizations that show branching paths, layered uncertainty, and scenario spaces instead of a single confident line. Useful framing for anyone building charts around forecasts, simulations, or counterfactuals.
Alberto Cairo |
| | |
| | Chartability is a practical audit framework for finding accessibility barriers in visualizations, dashboards, and data interfaces. Includes tests for contrast, keyboard navigation, screen readers, cognitive load, interaction design, and more.
Chartability |
| | |
| | | This site turns browser fingerprinting into a small, unsettling experience that reveals what your device discloses through ordinary web APIs. It’s more data than most people expect and is a sharp reminder that sharing data isn’t always opt-in.
since you arrived |
| | |
| | | This week's issue sponsored by: | | |
|
