| | This 100+ page guide covers everything you need to know about Airflow 3. New features, code examples, and your full upgrade playbook all in one place. Learn how to structure workflows, manage dependencies, and build pipelines designed to scale.
*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-570.html |
| | |
| | The Stanford AI Index is one of the most comprehensive, data-driven looks at the current state of AI. It pulls together benchmarks, economic signals, and public sentiment into a single view of how the field is evolving. Start with the Highlights section for the key takeaways.
Stanford Institute for Human-Centered AI (HAI) |
| | |
| | | Shazam doesn't match songs. It matches patterns. This post shows how Shazam strips audio down to a small set of stable features and then uses those features to search a massive index. It's a good example of how less data can work better.
Shri Khalpada |
| | |
| | Residual connections don’t forget, they just blur everything together. Attention Residuals swap that accumulation for learned routing across layers. In this post, Evan Lin reimplements Attention Residuals from scratch and offers a sharp tour of what PyTorch is actually doing under the hood.
Evan Lin |
| | |
| | If LLMs can generate the code, what’s left for you? In this keynote, Vicki Boykis argues the answer is everything that actually matters, and none of it comes quickly: data sense, system design, and the hard-earned intuition to know when things are wrong.
Vicki Boykis |
| | |
| | Most people prep for SQL and Python, but that’s not the cause of the rejection. DataDriven is a free platform that focuses on all parts of the loop, including data modeling, pipeline design, and defending decisions under pressure. It’s built around real interview loops, not practice drills, so you develop the intuition interviewers expect.
// sponsored |
| | |
| | | | What if your SQL query returned a chart instead of a table? ggsql adds visualization clauses directly to SQL, turning analysis and plotting into a single step. This is early, but the workflow shift is interesting.
Thomas Lin Pedersen, Teun Van den Brand, George Stagg, Hadley Wickham |
| | |
| | In her latest post, Nicola Rennie highlights lesser-known ggplot2 functions that make charts cleaner and easier to control, like overlapping axis labels, inconsistent text sizing, and awkward title alignment.
Nicola Rennie |
| | |
| | | If you’ve seen the RLHF Book, this is the next step. It’s now a full course, turning post-training into something you can actually work through, from high-level intuition to implementation.
Nathan Lambert |
| | |
| | | Josh Wolk turned NYC subway data into a live jazz performance where each line is an instrument, the city sets the rhythm, and every train is a note. It’s a simple idea, but the execution is kind of genius.
Josh Wolk |
| | |
| | | This week's issue sponsored by: | | |
|
