| | With so many new frontier models arriving, it's getting harder to know which ones are best at real work. Mercor's APEX-Agents leaderboard compares how well they handle complex assignments in banking, consulting, and law. Google's Gemini 4 models currently lead, ahead of Sonnet 5.5 and Opus 5.5. Explore the leaderboard. *Message from this week's sponsor, Mercor |
| | |
| | 🔗 Note: if you don’t see links in this email, it’s also available on the web: dataelixir.com/archive/issue-596.html |
| | |
| | Public comments don't always match true sentiment, but the U.S. is clearly at a fever pitch of data center opposition. Even if you’re not interested in the details, this is worth scrollling through. American Sentiment |
| | |
| | | This is great data detective work that starts with a minute of audio, a timestamp, and an aircraft you can’t hear. Can you figure out where it was recorded? YouTube | @colsto - 29 minutes |
| | |
| | Google ran an ad during the World Series that claimed baseball games get wilder during a full moon. And they had stats to prove it. Marcos Huerta checked the numbers and found a great case study in small samples, misleading percentages, and cherry-picked results. This is a fun investigation, including historical data, simulations, and lots of charts along the way. Marcos Huerta |
| | |
| | Filling missing values with a single best guess can make your statistical estimates look more certain than they really are. Multiple imputation offers a better approach by creating several plausible versions of the missing data to account for uncertainty. This hands-on walkthrough builds MICE from scratch in R and compares the results with an established implementation. Ken Koon Wong |
| | |
| | A model can predict which customers will cancel and still be terrible at identifying which customers a retention offer could save. That's one of many lessons in this hands-on comparison of causal ML methods and traditional models, using simulations where the true effects are known. Philipp Probst |
| | |
| | Every wrong answer from an AI agent is an opportunity to make the system better. In this webinar from AtScale, you'll learn how to capture failures as regression tests, update shared business definitions, and check future answers against known results. It's a practical approach to building AI analytics you can trust. Join the webinar on November 18. // sponsored |
| | |
| | | | This collection of curated articles goes beyond AI agent demos to show what data teams are actually building and what it takes to make those agents useful in practice. You can browse the articles by the type of problem they solve, which makes it a great resource for anyone working through similar challenges. GitHub | Cassis |
| | |
| | | This hands-on R tutorial combines ggplot2, ggiraph, htmltools, and Quarto to create self-contained interactive charts. It covers hover tooltips, dropdown filters, and a little bit of JavaScript to connect everything. The final result is an HTML file that anyone can open locally without a server. Nicola Rennie |
| | |
| | Do bigger dogs live where there’s more room? Mostly not. Starting with a simple question and NYC dog license data, this walkthrough uses maps and plots to work through missing data, a shaky proxy for size, and competing explanations before finding that income is a much stronger predictor than space. Kieran Healy |
| | |
| | | A Klein bottle is a mathematical surface that loops back through itself, with no distinct inside or outside. It turns out that it also describes a centuries-old paradox in voting theory. This article shows how researchers uncovered the connection, with visualizations that make the abstract mathematics easy to follow. Quanta Magazine | Alex Stone |
| | |
| | | This week's issue sponsored by: | | |
|