Hello, reader!

Quantum #649 had an open rate of 51% and a click rate of 15%.

The most clicked link was this beautiful visualization of lighthouses.

'till next week,
Giuseppe

Topical

For Nordics, modern-day ice age could be close

The European Correspondent on the possible "collapse of the Atlantic Ocean current system could significantly cool Northern Europe."
The Atlantic Meridional Overturning Circulation (AMOC) is the ocean current system regulating Europe's climate. It has been weakening over the past few decades, and models suggest it might reach a tipping point within 10-20 years, which would make it shut down 50-100 years later.

Jane Street depends on all sorts of messy, real-world data to understand financial markets and the global economy: think world news, decades of weather patterns, deidentified credit card spending, or packet captures of stock exchange market data feeds.

We're hiring Data Engineers to turn datasets like these into reliable inputs for trading. Working closely with our researchers, you'll evaluate unfamiliar datasets, build robust ELT pipelines, develop deep domain expertise, and decide what's worth exploring next. 

The job requires a mix of engineering, data analysis, and product sense. If you love the detective work of investigating a weird dataset and figuring out what it actually means, we want to hear from you. No financial background is necessary.

We have openings in New York, London, and Hong Kong.

Tools & Tutorials

Night Shift Agentic Workflow

Jamon Holmgren describes his agentic AI workflow: "I decided that I will take the day shift, and AI agents will take the night shift. I will prepare everything for them as much as I can during the day shift. Then, during the night, they will work autonomously while I am resting, and be done by the next morning."
Fundamentally, this deals with using AI in a light-touch way for requirement specifications during the day, then deploying AI agents to work autonomously during the night. The article also suggests some prompts for how to use agents for code reviews. We'll see more of this kind of idea.

a5R

Pentagonal geospatial index library A5 has been ported to R.

How to convert Excel spreadsheets to Python models with Claude Code - Martin Alderson

This is a five-step process for converting complex Excel financial models into Python modules using Claude Code.

Data Wrangling in Stata

This online book and JupyterLab app teaches data wrangling skills in Stata, a popular data analysis tool in the social sciences.

Coding agents for data analysis

Simon Willison ran a three-hour workshop at NICAR 2026 teaching data journalists how to use AI coding agents like Claude Code and OpenAI Codex for data exploration, analysis, and cleaning. All the contents are available here.

worldmonitor

"Real-time global intelligence dashboard — AI-powered news aggregation, geopolitical monitoring, and infrastructure tracking in a unified situational awareness interface." It's available as a self-hosted open source package, too.

How to find data

A slide deck from a presentation held at the recent NICAR conference, a yearly gathering about computational journalism.

NativePHP/mobile-air

NativePHP is a framework to user PHP for desktop and mobile apps. This is specifically the mobile package. NativePHP is based on Laravel and MIT-Licensed.

CanIRun.ai

CanIRun.ai is a tool that helps users determine whether their computer hardware can run various AI language models locally. You can use it in two ways: automatically, by detecting GPU, VRAM, bandwidth, RAM, and CPU cores through browser APIs; or manually, by selecting your machine from a drop-down list.
Models are graded on a performance scale from "Runs great" to "Too heavy" and can be filtered by task type (chat, code, reasoning, vision), etc.

JPEG compression

"JPEG is a clever image compression algorithm that exploits a human perceptual bias and the structure of natural images. It makes a change of basis before encoding so that the rewritten image representation concentrates the signal that humans are most perceptive to, and thus, the remaining details can be discarded without horribly degrading visual fidelity."

How to Estimate a Mean, and What It Means for Science

"If I asked you to estimate 30 means, you would probably compute 30 sample means... once we are estimating three or more means simultaneously, the sample mean is inadmissible—meaning there exist other estimators that have lower total mean squared error for every possible set of true values."

syntaqlite: high-fidelity devtools that SQLite deserves

Lalit Maganti's syntaqlite is an "open-source parser, formatter, validator, and LSP built directly on SQLite's Lemon-generated grammar". This ensures it interprets SQL exactly as SQLite does across different versions and compile flags.

claudetop - htop for your Claude Code sessions — real-time cost, cache efficiency, model comparison, and smart alerts

claudetop is "htop for your Claude Code sessions — real-time cost, cache efficiency, model comparison, and smart alerts".

"I built claudetop after noticing my model estimate showed $10 but the actual bill was $65. Turns out, compaction was hiding 80% of my token usage. The cost was real — the visibility wasn't."

Environmental Impact of Airports – Wrestling with Data

Steven Feldman attempted to create a map showing airports and their environmental impact. The resulting interactive map visualizes estimated emissions on a color scale and can display data by passenger numbers or population impact, with a year slider and zoom-responsive statistics.
"Some smaller airports can have a bigger impact because they are located close to large populations eg London City Airport is a small airport with about 3.5m passengers n 2024 and estimated CO2 of under 0.5m tonnes but because of it's location it has a moderate impact on 3.7m people while Heathrow with 84m passengers and 10.5m tonnes of CO2 is more rural and only impacts 1.6m people."

Data Thinking

When building becomes cheaper than specifying, everything changes

"How the cost of building collapsed, and what it means for product teams and companies."

Dataviz, Data Analysis, & Interactive

25 Years of Eggs

John Rush collected receipts of egg buying since 2001 , scanned them, and used AI to extract the data. We're talking about 11,345 receipts.
This article is interesting on various accounts: the sheer absurdity of the task, the monumental size of it, and the technical details of how traditional methods failed.

(via Web Curios)

London's Divide Was Called Character. It Was Actually Policy.

Lauren Leek: "I built a machine learning model to find London's divide and you can enter your postcode to see which side you're on. We've been blaming the wrong people for it". She basically looked at neighbourhood-level data (education, occupation, ethnicity, house prices, café and pub density) to identify London's socioeconomic divide. The model achieved 95% accuracy in classifying areas, aligning with the east-west split that Charles Booth famously suggested in 1889. Other than the article, play with the interactive map.

The Billion-Dollar Business of ABBA

Daniel Parris (Stat Significant) looks at "how ABBA turned joy, nostalgia, and musical IP into a billion-dollar enterprise."

US Job Market Visualizer

Andrej Karpathy has released this open source research tool, obviously created with LLMs, to assess AI-powered transformation of 342 occupations listed in the US Bureau of Labor Statistics.
"A high score does not predict the job will disappear. Software developers score 9/10 because AI is transforming their work — but demand for software could easily grow as each developer becomes more productive."

AI

LLM Architecture Gallery

A nicely illustrated gallery of large language model architectures, documenting models from GPT-2 (2019). Each entry comes with configuration files, technical reports, licensing information, and key architectural details.

aiglossary.news

"A free, open-source glossary of 51 AI terms and concepts for journalists,", part of a session at NICAR.

Measuring US workers' capacity to adapt to AI-driven job displacement

This Brookings research combines AI job exposure estimates with a novel "adaptive capacity" measure, trying to account for workers' savings, age, labor market density, and skill transferability.

Startup Technical Guide – AI Agents (PDF)

Google's PDF guide for startups wanting to adopt agents. Obviously based on Gemini.

Other

Visualizing Ukkonen's Suffix Tree Algorithm

After facing the difficulty of learning complex algorithms from textbooks, Alessandro Bahgat decided to visualise them. This blog post focuses on Ukkonen's suffix tree algorithm.
"The biggest frustration was that I had no way to inspect what my code was actually building. I relied on the usual bag of tricks: print statements, breakpoints, inspecting memory structures one by one in a debugger. But that is like understanding a forest by looking at one tree at a time. What I wanted was to see the whole data structure after each operation — to watch the algorithm work."

AMA – Ask Me Anything! Submit a question via this anonymous Google form. I'll select a few every 4-5 weeks and answer them on here :-) Don't be shy!

The Quantum of Sollazzo grove now has 40 trees. It helps managing this newsletter's carbon footprint. Check it out at Trees for Life.

'till next week,
Giuseppe @puntofisso.bsky.social