| | Marc Lamberti is hosting a live AI Orchestration crash course at Orchestrate Everything, the virtual data engineering conference on Sept 16. Learn to run and evaluate production AI workflows in Airflow and get a free exam code ($150 value) for the official certification. Then, see what production AI orchestration actually looks like with sessions from builders at Lyft, Ramp, and Wix. Register now
*Message from this week's sponsor, Astronomer |
|
| | 🔗 Note: if you don’t see links in this email, it’s also available on the web: dataelixir.com/archive/issue-587.html |
| | |
| | Retraining a struggling model isn’t just one decision. Airbnb argues it’s three: refit the parameters, change the model itself, or leave it alone. This post shows how to tell the difference, using examples from Airbnb’s forecasting models during and after the pandemic.
Airbnb Tech Blog | Harrison Katz |
| | |
| | Foundation models lag traditional forecasting methods in terms of accuracy, but they’ve created a real shift in how forecasts get produced. In this post, Rob Hyndman breaks down the capabilities of foundation models in 2026 and offers practical advice for thinking about where they fit in modern forecasting.
Rob J Hyndman |
| | |
| | ACF plots are easy to calculate but can be tricky to interpret. This post shows how to read the pattern across lags, why confidence bands can mislead, and how trend or non-stationarity can make a series look far more persistent than it really is.
M. Fatih Tüzen |
| | |
| | In this post, the team at Data Colada shows how a handful of basic data checks can reveal a dataset that doesn’t behave the way real data should. The initial tip was a replication failure of a well-known experiment.
Data Colada | Uri, Joe, & Leif |
| | |
| | Bayesian experiments are often described as faster, but the statistics alone don’t make that true. The real advantage is that Bayes gives you different ways to decide whether collecting more data is actually worth it.
Demetri Pananos Ph.D |
| | |
| | RAG doesn’t have to mean embeddings and a vector database. In this post, Rafael Pierre lays out a progression of six retrieval approaches that add complexity as your needs evolve.
Lighthouse AI | Rafael Pierre |
| | |
| | | | LatticeDB is essentially a SQLite-style database for graphs. It runs locally in a single file and combines graph queries, vector search, and full-text search without requiring separate services.
GitHub | Jeff Hajewski |
| | |
| | | The butterfly effect is sometimes used to describe how a tiny change can send a big system down a completely different path. It's mostly used as a metaphor but this interactive traces ten real-world cases that had small beginnings that led to very large consequences.
Akash Wadhwani |
| | |
| | | This week's issue sponsored by: | | |
|
