Things I've been reading this week

A new digest

Every week, I read lots of fascinating things on Substack, and often post snippets on Notes. But I thought I’d try something new, and instead compile them into a weekly summary, in case you’ve missed any of them and fancy having a look.

So here are a dozen interesting things I’ve come across this week. It doesn’t mean I agree with or endorse every word of every article. But they all sparked some thinking that I found valuable.

I’d love to hear what you think in the comments – is this kind of digest useful? Would you prefer weekly, month etc? And read to the end for news about an exciting new project!

Semantic Apocalypse Now by James Taylor Foreman

if perfect chess robots don’t make Magnus Carlsen irrelevant, I struggle to see how LLM’s are suddenly going to start “solving” reality more generally just because they “solve” the simplified game we created to represent it

How do you solve a problem like the high street? by Polly Mackenzie

Landlords would often rather have a vacant building with a nominal rental value of £10,000 a month than a full building rented out at £5,000 a month. Lower rent may mean cash in the bank but it forces them to realise a loss in the capital value of the building.

Saloni’s guide to data visualization by Saloni Dattani

When I did data visualization during my PhD, I received no training in it. I think it’s a skill that’s often seen as a nice bonus, or simply difficult to do well and better left to other professionals. I hope I’ve convinced you of the opposite.

How did we get a regression model to the top of the Virtual Cell Challenge by Kris Szalay

So don’t dismiss simple models like regression or Random Forest. Given the same data, they can perform on par with far more complex architectures, and often generalize better to other tasks.

Is it really a super flu year? by Christina Pagel

The important question is whether this year’s flu season will peak soon (and so we have a serious flu wave but a ‘normal’ serious flu wave, just shifted a few weeks earlier) or whether it will keep going up and only peak at towards the end of the year in which case it will be a terrible flu season - i.e. are we in scenario A or scenario B in the chart below. We don’t know! The data simply doesn’t tell us that!

How to party like an AI researcher by Jasmine Sun

It’s a funny sight: hundreds of people interviewing for $500k/year roles, willing to wait 30 minutes backpack-to-backpack to save $5 on a cappuccino.

Pretraining: The First Scaling Frontier by Bauer LeSavage & Simon Barnett

When originally presented, the concept of emergence brought significant concern to the field as downstream performance felt unpredictable. However, elegant work by Schaeffer et al. showcased there may actually be nothing to worry about. Instead, “emergence” is the result of poor experimental design and execution – selecting nonlinear (% accuracy) or discontinuous (multiple choice) metrics with limited data points can trick researchers into mistaking low-resolution graphs for breakthrough results.

How Renaissance Italy fought plague by Ming Yang

After a century of devastating epidemics, including the Black Death, and with trade routes bringing both wealth and infection, Italian cities realised that survival required a standing system capable of spotting threats early and acting fast. By the mid-1400s, many of them, led by Venice, created something unprecedented: permanent health boards (magistrati di sanità) with full-time staff, budgets and legal authority.

Thank God for Bluesky by Ian Dunt

I am therefore responsible for the fact that it is not providing a broad enough range of experience. Or perhaps other users are for not writing enough of that content, or for not sharing it enough. Either way, it’s on us. We are building our own experience from the bottom up, by our choices.

We are in the era of Science Slop by Jonathan Oppenheim

There’s just one problem: the paper answers a question whose answer we’ve known for 35 years.

Let’s Not Bring Back The Gatekeepers by Dan Williams

people—well, not scientists or professional journalists or highly-educated professionals who read broadsheet newspapers and believe in truth and reason and facts and evidence, but everyone else—are irrational, emotional, and stupid, credulous towards misinformation and yet pig-headed in the face of evidence-based arguments

Mind the AI gap by Oliver Morgan

The surveys show that many organisations adopt AI but hold back from the deeper changes that create value. Tools are not the barrier, it is the underlying systems. Public health agencies are no different. Data quality, workflow design, and staff capability determine whether AI adds value.

From long reads to medium clips

Finally, this week was the launch of a new video series I’ve made with the support of the TED Fellows program. Much like this Substack, the videos explore how we can make sense of a noisy, uncertain world. What is really happening? What should we do about it? And how can we communicate these things to others? These initial videos give a short, sharp video overview in 4-6 mins (rather than a longer Substack post). But would love to hear what you would find most valuable. Longer deep dives? More frequent short clips? Let me know!

I’ve also learned a huge amount about the magic tricks of online video while working on this project, so will post about it in the future – a bit like my previous post on the unusual lessons I’ve picked up across hundreds of media interviews: