| | Data agents aren't the problem. Bad data is. This blueprint shows you how to build a semantic layer on Snowplow data in Snowflake and connect it to Snowflake CoWork, so your teams get accurate answers instead of confident guesses.
*Message from this week's sponsor, Snowplow |
| | |
| | 🔗 Note: if you don’t see links in this email, it’s also available on the web: dataelixir.com/archive/issue-583.html |
| | |
| | A model can give the right answer, explain it clearly, and still be reasoning for the wrong reasons. This article explores the growing evidence that “reasoning traces” may be less trustworthy than they seem.
Quanta Magazine | John Pavlus |
| | |
| | A lot can hide in summary statistics. This post shows how visualization techniques like CDFs, ridgelines, heatmaps, and segmented analysis can reveal the stories that averages hide.
Farid Zakaria |
| | |
| | This is a great hands-on tutorial that shows how to use and evaluate local multimodal models. "Gymflation" is a fun idea that uses data from YouTube to build, test, and improve a local-model analysis.
Chris Bowdon |
| | |
| | There are 3,170 pubs in London, and not one of them is within an hour of all 33 boroughs. This post uses geospatial analysis, TfL journey data, and optimization thinking to find the fairest place for a group to meet. There are 2 clear winners, and neither is in the center of London.
Equidistance |
| | |
| | Hex is an analytics platform where analysts do deep technical work, business users self-serve in plain language, and every answer is grounded in the same shared context. Data Elixir readers get 30 days free.
// sponsored |
| | |
| | | | This open book is a practical guide for developing predictive models from tabular data. It's a work-in-progress but there's a lot here to get started with. Covers data splitting, preprocessing, feature engineering, resampling, tuning, model comparison, and classification.
Max Kuhn and Kjell Johnson |
| | |
| | | Paste in some Python code and this tool identifies the earliest compatible version. It runs in your browser and doubles as a fun tour through Python’s evolution.
Python Morsels |
| | |
| | smevals makes it easy to compare models, prompts, and agent harnesses on your own specific tasks. The results are also reproducible, with graders you can refine over time and use to reevaluate earlier runs.
Prime Radiant | Jesse Vincent |
| | |
| | | Lightning strikes can be detected and precisely located across the entire planet, in real-time, and from thousands of kilometers away. Here's how advanced math and sensors make it work.
Counting Stuff | Randy Au |
| | |
| | | This week's issue sponsored by: | | |
|
