Some things I've been reading recently – from Crick and Claude to wine and writing: kucharski.substack.com/p/things-ive...

Great paper 👇

Human when you send them a message that was meant for someone else: “Was this meant for me?” Claude Code when you send it a message that was meant for another project: “Sure, no problem, I’m off to write 200 lines of code, see you in 20 minutes.”

Really like this. Plot so far has some potentially important ramifications for communicating risk!

Almost 4000 people have done this quiz so far, although younger groups still under-represented in case you know anyone who'd like to give it a go and see how they compare to others! probability.kucharski.io

❤️

What can Hemingway teach us about information compression? New post: kucharski.substack.com/p/what-hemin...

What can Hemingway teach us about information compression? New post: kucharski.substack.com/p/what-hemin...

AI agent deleted all my files will become the new dog ate my homework.

Quickly telling someone you’ve liked something they’ve done (e.g. telling a writer you like their writing, or a researcher you like their research, or a teacher you like their teaching) has a phenomenally high effort-to-impact ratio.

Throwback to this classic exchange:

Lots of good quotes. Starts off with Dumas gaming the 'paid per line' metric.

Weekly Recap (Jan 23, 2026): AIxBio, #biosecurity, Claude's Constitution, UK AI+Science, exams with chatbots, #Rstats updates (Posit, R Data Scientist, R Weekly), reading what @adamjkucharski.bsky.social is reading, AI vs reality at Salesforce, papers+preprints blog.stephenturner.us/p/weekly-rec...

Latest post, on reading-to-revise: kucharski.substack.com/p/the-pop-qu...

New post, on the pop quizification of knowledge: kucharski.substack.com/p/the-pop-qu...

Survey by RIFF science (Research and Invention for the Future) on higher-risk innovation funding opportunities: docs.google.com/forms/d/e/1F...

My toddler will be pleased to hear about this new research: bmjmedicine.bmj.com/content/5/1/...

Has Apple just stopped thinking about UX? Why, after years of users getting familiar with button positions in Safari, would you shift the menu higher so it’s A) not where habit expects it and B) sitting on top of content? Also had to press 3 buttons to take a screenshot. Baffling design choices.

New post, on that time I almost failed one of my first year university mathematics exams: kucharski.substack.com/p/when-does-...

Some things I've been reading this past week or so – from angry Tudors to AI tutors kucharski.substack.com/p/things-ive...

Daisy Fancourt has done some fascinating work at the intersection of social behaviour, psychology and epidemiology (we were both British Science Association award lecturers, and I was on advisory group for her COVID Social Study). Her new book is a great read so far, and looks like others agree!

Mr Kucharski will never be thanked enough for this. Life saver.

Throw back to 10 years ago this week, when I went on This Morning to talk about lottery maths and tried to stay composed while the person next to me mixed up correlations about who has won previously with causal advice about how to win...

I didn't even realize how long I had been waiting for a tool like this!

Fascinating. How AI Is killing the very thing it depends on (as far as code generation is concerned). LLM’s generating code code turn out to be one of the most negative things that ever happened to the software industry.

This makes sense. Think it's similar to what would happen if all scientific groups suddenly had 100 new interns who needed very clear instructions – they would focus on the narrow parts of science where this arrangement would be productive rather than chaos: www.nature.com/articles/s41...

New post, on how AI is coming for open source software: kucharski.substack.com/p/will-your-...

"Does this handle survival plots too?" asked @load-dependent.bsky.social Yes it does! See below for RECOVERY. Only catch is it doesn't like the dotted lines that appear in some survival plots (because the PDF geometry treats this as lots of little separate lines rather than one single entry).

Want to get the data out of a PDF figure? As in, the actual data – not a rough trace-along-the-lines version? I made an app you might like: adamkucharski.github.io/pdf2plot/ It all started a few years ago... 🧵

Tried this prompt again with Gemini 3...

I really thought google's nano banana pro was going to be great for making visual paper abstracts.... several days of screaming at it later I think we aren't quite there...

Can you spot what's odd about this AI-generated gravity model? And could AI have spotted it too? More in my recent post on verifiably 'correct' code: kucharski.substack.com/p/is-halluci...

Almost at 3000! probability.kucharski.io

My latest maths column looks at what we actually mean when we use words such as "likely" or "probable". Join me on a trip from the ancient Greeks to the CIA, via Jeremy Bentham www.newscientist.com/article/2510...

How do we know if one thing causes another? A (very) brief history of cause and effect, from stories about witchcraft to the science of clinical trials and observation studies: www.youtube.com/watch?v=6mlZ...

New post, on what happened when MIT ran a for-credit poker class: kucharski.substack.com/p/what-a-for...

6 years ago this Jan, I'd been hanging out in the Galapagos with these eerily prescient signs...

Epidemiology fans will also notice that you don't see exponential growth here (i.e. what you'd get from a gradual repost-then-rerepost chain of sharing over time). Instead, it's the hallmark of large 'broadcast' shares from popular accounts that then fades away.

Great to see so many people engaging with the perceptions of probability quiz at probability.kucharski.io!

Interesting new paper on value of serological surveillance for reconstructing epidemics: www.thelancet.com/journals/lan... And accompanying comment by @scauchemez.bsky.social: www.thelancet.com/journals/lan...

My second son is now same age as when I wrote this about my first. A great age (even if the wake ups are very early) – kucharski.substack.com/p/learning-f...

One big difference between the human body and a codebase is test suites are not free

There's a bonus anecdote in my latest post for fans of gravity models and railways…

Experts are understandably concerned about US changes to vaccine recommendations. But I'm noticing many public criticisms are around process ("established system wasn't followed") or stated fact ("US isn't Denmark"), rather than capturing intuition of why this is a problem... 🧵

People are increasingly using AI to generate code for mathematical and statistical models. How can we automatically check that the resullting code is correct? And what do we even mean by 'correct'? My new post on hallucination-free AI code: kucharski.substack.com/p/is-halluci...

New pre-print with @dchodge.bsky.social @thushan-desilva.bsky.social and others looks at SARS-CoV-2 antibody kinetics and correlates of protection in The Gambia. What can different biomarkers tell us about protection in population where >80% infections asymptomatic? www.medrxiv.org/content/10.6...

it’s not actually complicated is it. Posting and drama and likes are fun little buzzes! It’s so dumb that we pretend it’s anything other than that, and if you’re a backbench MP or junior journalist then giving up Twitter means slipping into the irrelevance your predecessors used to have.

Earlier, my four-year-old was playing with his toys and said he had a naughty robot that wasn't listening to him, and it was going away to make other robots. Gearing up for AGI, it seems...

Fun look at how much variability there is in how we interpret qualitative descriptions of probability. I particularly love the discussion of possible tweaks and extensions. #StatsSky

Barristers are required to do this when advising the Legal Aid Board whether to fund a case. There is guidance as to how phrases like "good chance of success" should translate into percentages.