| | If your dashboards disagree or your AI outputs feel off, the issue might be upstream. This summit looks at how teams are using semantic layers to define metrics once, share context across systems, and avoid things drifting as AI gets involved.
*Message from this week's sponsor, AtScale |
| | |
| | Note: Data Elixir is taking next week off and will be back in your Inbox on April 28th. Have a great couple weeks! |
| | |
| | Math has been one of the hardest domains for AI but that’s starting to change. This article tracks the shift from toy experiments to systems that can explore, test, and refine ideas at scale, hinting at a new kind of research workflow.
Quanta Magazine | Konstantin Kakaes |
| | |
| | | The hardest part of building AI systems isn’t the model, it’s everything around it. This post argues that evals, metrics, and data inspection are the real bottleneck, and they look a lot like data science.
Hamel Husain |
| | |
| | Big models don’t have to stay big. This guide walks through quantization from first principles, showing how you can shrink models, speed them up, and still keep most of their performance.
ngrok blog | Sam Rose |
| | |
| | | Average speed hides everything. This post breaks bus routes into segments and shows how performance shifts block by block, pulling together route shapes, stops, and time-based speed data.
MTA Data and Analytics | Emmalie Morisseau and Benjamin Tannenwald |
| | |
| | You can’t “model your way” to causality. In this post, Frank Harrell explores how causal inference is used in practice, and offers a reminder that design, measurement, and domain knowledge matter more than any framework.
Frank Harrell |
| | |
| | | This collection of Jupyter notebooks rebuilds core ML algorithms from scratch and shows how they learn, so things like convergence and loss landscapes are much easier to see.
Gavin Hung |
| | |
| | This reads like the stats course most people wish they had. One coherent path from core probability to modern Bayesian workflows, with consistent notation, real computation, and fewer gaps between concepts.
Carlisle Rainey |
| | |
| | | Take a familiar system, remove almost all degrees of freedom, and see what survives. This 1D chess variant turns the game into a constrained search problem where optimal play quickly becomes the whole story.
Rowan Monk |
| | |
| | | This week's issue sponsored by: | | |
|
