Data Science Weekly - Issue 663

Data Science Weekly Newsletter

SubscribeSign in

Data Science Weekly - Issue 663<br>Curated news, articles and jobs related to Data Science, AI, & Machine Learning<br>Data Science Weekly<br>Aug 06, 2026

Share

Issue #663<br>Augu 23, 2026

Hello!<br>Once a week, we write this email to share the links we thought were worth sharing in the Data Science, ML, AI, Data Visualization, and ML/Data Engineering worlds.

Sponsor Message<br>3 Enterprise AI Breakthroughs.<br>One Live Event.

See how to build, own and compound your AI on 8/11.

General-purpose frontier AI is built to be good at everything. It<br>wasn’t trained on the tasks that differentiate your business, and it<br>stops learning about your business the day it ships.

Join the Oumi team on August 11 for the latest addition to the<br>industry’s most powerful AI factory. See how to build, own, and<br>compound AI that knows your business, improves with every production<br>run, and becomes an advantage no competitor can rent.

What you’ll learn:<br>Why building, owning, and compounding is the right strategy: where<br>specialized AI beats frontier models on quality, control, and cost,<br>and where it doesn’t.

How to run a complete cycle on a real task: Evaluate, Synthesize,<br>Train, Deploy, Compound.

Three announcements we’re only revealing live. One changes how<br>enterprise agents get built. One kicks in when your model hits<br>production. One puts the AI Factory where your developers already<br>work.

Your AI stopped learning the day it shipped. On August 11 at 10 AM PT,<br>see what it looks like when it never stops.

Can’t make it live? Register anyway and we’ll send the recording.

Join Live

And now…let’s dive into some interesting links from this week.

Editor's Picks

A Primer to ML Compilers<br>This presentation was given at an ML systems reading group at CMU over the summer. I have taken the slides and added notes underneath to share the presentation. This largely covers the papers that introduced TensorFlow, PyTorch 2, and TVM, along with the preceding works that laid the foundation or inspired these projects. This presentation and its notes should provide a fundamental overview of the ML compiler landscape to get started…

How Do I Build a Model? A starting guide for new modellers<br>This guide is for the graduate student who is beginning to work in modelling and has asked: how do I actually build a model? I’ll assume you already know roughly what your research question is and that you have some goal in mind. If you don’t yet, that’s the first thing to sort out. But the heart of the guide is what to do once you have a question and want to turn it into a model…What I share here is a guide to meta-principles for how to start with model building. They apply to any type of modelling. Once you get to learning about a specific type of modelling then I recommend you read a book by a discipline expert. There’s a short reading list at the end that covers my favourites…

measuring scientific progress<br>I’ve measured scientific progress by computing the lifetimes of 3,444 scientific claims. I made Kaplan-Meier curves showing how long it takes for a claim to go from appearing in a review paper to its demise. Claims made in the 1950s have an extrapolated half-life of 250 years, while those from the 2000s have a half-life of 70 years. If this acceleration continues, scientific claims will live only 1 year by 2212. That would feel like a scientific singularity, whereby the scientific record is significantly revised each day…

What’s on your mind

This Week’s Poll:

Loading...

Last Week’s Poll:

Data Science Articles & Videos

Retire the Abstractions - We think CUDA DSLs are heading toward retirement<br>Last year, we set out to write megakernels. But implementing megakernels can be quite miserable. It requires complex data structures, extensive inter-thread/SM/GPU synchronization, deeply nested control flow, etc. We couldn’t hold it in our heads, so we did what computer science has done for 70 years: we built a layer of abstraction. Even with it, we had to fight through race conditions and deadlocks for a couple months before the Llamas ran blazingly fast. This year, we built an MoE megakernel, but we deleted the abstraction. With agents, we could work through the complexity directly and build target-optimized code from scratch. No intermediate layers of C++ abstraction were needed. So what could this mean?…

Power 2026 - Electricity Pricing in the Age of AI<br>I’m a former quant researcher at a major hedge fund who covered power and gas, and I’ve spent much of the last couple of years advising founders and investors on data center buildouts (procuring GPUs, negotiating with coal plants, identifying sites)…This primer on power and data centers is for a broader audience who senses there’s an opportunity in energy markets and wants to get up to speed…By the end of this primer, you’ll understand:<br>Part 1: how power plants work , how data centers are developed , and how companies might respond (for...