| | We get it—you want the freedom to explore new tech, and you want a solid foundation for job growth. Well at Fidelity, that’s not a trade off. As a privately-held company, we invest in the resources that matter most for the long term. This means more opportunities and peace of mind for our technologists to build, learn, and invent for the future. Learn more about tech at Fidelity.
*Message from this week's sponsor, Fidelity Careers |
| | |
| | 🔗 Note: if you don’t see links in this email, it’s also available on the web: dataelixir.com/archive/issue-590.html |
| | |
| | This is a good data-wrangling gotcha with real statistical consequences. In this post, Danielle Navarro shows why dplyr::ntile() can split identical values across different groups, and why its SQL-style approach to bucketing isn’t always appropriate for statistical analysis.
Danielle Navarro |
| | |
| | AI is making it easier to generate correct answers, which makes it more and more important to be able to explain and judge them. The vehicle here is mathematics, but the point is broader. Producing an answer is different from understanding where it came from and why it works.
Terry Tao Blog | guest post by Grant Sanderson (3Blue1Brown) |
| | |
| | How do 288 Sudoku solutions turn into just 12? In this post, Francesco Baldino uses a tiny 4×4 puzzle to explore brute-force search, symmetry, and a neat bitmask trick for finding minimal subsets. It’s a compact example of how much you can learn from a deliberately small problem.
Francesco Baldino |
| | |
| | ROC-AUC is useful, but it only tells part of the story. In this post, Ken Koon Wong works through discrimination, calibration, Brier scores, and decision curve analysis to show why a model that ranks well still may not give probabilities you can trust or decisions worth acting on.
Ken Koon Wong |
| | |
| | There are plenty of benchmarks that show what AI models can do. Mercor’s APEX suite is more interested in whether they can do real professional work. It tests models and agents on tasks from banking, law, consulting, accounting, medicine, and software engineering, with leaderboards that update as new frontier models ship. See the benchmarks →
// sponsored |
| | |
| | | | AI agents can build dashboards with React, JavaScript, and charting libraries, but that still leaves a lot of code to review and maintain. dbt Charts is a new open-source format that takes the opposite approach. With dbt Charts, dashboards are described in a simple YAML file, which makes a dashboard easier for agents to modify and for humans to review.
dbt Charts |
| | |
| | | PostgreSQL 19 is getting close to release, and this is a hands-on tour of what’s changing. Jesús Espino works through new features with runnable SQL examples, including better COPY support, temporal updates, concurrent table repacking, query-plan controls, and a long list of small improvements.
Jesús Espino |
| | |
| | Rig is an installation manager that lets you easily run and manage multiple versions of R side by side on macOS, Windows and Linux. The latest beta adds user-mode installs, package management for CRAN and Git packages, and tools for managing projects and virtual environments.
rig-r-lib |
| | |
| | | Charles Petzold wondered whether the spacing of guitar frets could be used for multiplication, so naturally he sawed a virtual guitar in half and turned it into a slide rule. It works through one octave, then breaks in the second. The reason it works is stranger than the reason it doesn’t.
Charles Petzold |
| | |
| | | This week's issue sponsored by: | | |
|
