Adam Kucharski

"It’s not difficult to say something interesting. It’s not difficult to say something true. The real challenge is saying something both interesting and true."

The Mr Beast strategy vs The Rolling Stones strategy Is getting online buzz worth the effort?

New post on the MrBeast strategy vs The Rolling Stones strategy when it comes to chasing online attention: kucharski.substack.com/p/the-mrbeas...

At 17, Hannah Cairo Solved a Major Math Mystery | Quanta Magazine After finding the homeschooling life confining, the teen petitioned her way into a graduate class at Berkeley, where she ended up disproving a 40-year-old conjecture.

Earlier this year, a 17-year-old high school student named Hannah Cairo solved a 40-year-old mystery about how waves behave, surprising and exciting mathematicians. @kevinhartnett.bsky.social reports: www.quantamagazine.org/at-17-hannah...

According to the UK govt probabilistic yardstick, which is designed to avoid woolly statements, 'realistic possibility' = 40-50% probability and 'highly likely' = 80-90%.

So, based on this scale, he seems to be claiming that, on balance, it wasn't a lab leak? But also he's very confident it is?

You know you’re getting old when you see the word ‘pipe cleaner’ and don’t think ‘thing to make stuff at school’ but instead think ‘thing to clean pipes’.

The Uncertain Science Behind What We Understand As ‘Truth’ In a new book mathematician Adam Kucharski traces our relationship to truth from the ancient Greeks to our AI reality.

In his new book "Proof," mathematician @adamjkucharski.bsky.social explores our complicated history of separating fact from fiction.

buff.ly/joJHPeh

Quite. People's revealed behaviour is usually at odds with their professed preference for the past...

Great analysis by @davelee.me that Google’s AI summary is a solution to a problem Google itself created: an ad-riddled, keyword-optimized web hostile to conveying useful information

The real-time infection hospitalisation and fatality risk across the COVID-19 pandemic in England - Nature Communications The severity of SARS-CoV-2 infection varied over the course of the pandemic due to factors such as changes in variant characteristics and population immunity from previous infection or vaccination. He...

Would have to dig around – REACT and ONS were used a lot in real-time, but death certificate data too delayed. There has been some follow up (and discussion of various data): www.nature.com/articles/s41...

Please, no more group brainstorming And other ways to make event content more valuable

"A common experience at the end of a brainstorming session is someone asking, “okay, who has capacity to do this?” The answer, dear reader, is absolutely nobody at any point in time."

Also, evaluation papers need to move away from this narrow focus on deaths as the only COVID harm worth considering - vaccines also prevented a LOT of hospital and ICU admissions, many of which would have had life altering consequences.

Intro to the new Ioannidis et al paper says 'Models may give unreliable results, depending on assumptions' – then it proceeds to use a model with very questionable assumptions, ignoring the best available IFR data...

Famine and genocide in Gaza: A personal view | LSHTM Famine and genocide in Gaza: A personal view | LSHTM

Thoughtful piece on the Gaza crisis by Francesco Checchi: www.lshtm.ac.uk/research/cen...

Useful new review of SARS-1 parameters: www.sciencedirect.com/science/arti...

Two PhD studentships available in real-time infectious disease modelling, jointly with LSHTM, Imperial and UKHSA. January 2026 start + 3.5 years of funding: www.lshtm.ac.uk/study/fees-a...

Excellent synopsis with a practical application

This is really fascinating. I like that it offers a framework for deciding annoying little decisions like whether or not to purchase things like travel insurance or extended warranties or more car insurance.

How much uncertainty is too much uncertainty? And does it matter where it comes from?

Suppose you flip a coin. If it comes up heads, I give you £20. If it comes up tails, you give me £10.

Would you take this bet?

What if the stake was £200,000 vs £100,000?

New post on how much uncertainty is too much uncertainty:

British boasts of being a ‘science superpower’ are a bit cringe Not since that other phrase, ‘follow the science’, has there been so much unease that words that sound like they are backing science might be a trap for it

"Not since that other phrase, “follow the science”, was popularised has there been so much unease that words that sound like they are backing science might actually be a trap for it." www.thetimes.com/article/42d5...

I also think the "quarantine failures per..." also speaks to the meta issue of a country's willingness to completely (and quickly) reshape the way it does everything.
NZ's 98% reduction in the level of people arriving (Q2&3 2020 vs. 2019) was scaling to the level of available quarantine not demand.

I agree. NZ and Aus geography also made border control easier to implement. In NZ there were only 2 points of entry for passengers (Chc and Akl airports) & shipping freight is at sea for weeks (>>1 incubation period) before arrival. So failures per 100,000 would likely have been higher in the UK too

🧵Introducing the Trump Action Tracker website!

Today I’m launching www.trumpactiontracker.info - a live, searchable list of authoritarian‑style actions from Trump’s second term (over 740 actions so far). 1/14

Today at 12.30pm UK time!

Failures of quarantine systems for preventing COVID‐19 outbreaks in Australia and New Zealand Objectives To identify COVID-19 quarantine system failures in Australia and New Zealand. Design, setting, participants Observational epidemiological study of travellers in managed quarantine in Au...

This is a useful paper, which quantifies quarantine failures per 100,000 travellers: onlinelibrary.wiley.com/doi/full/10....

I suspect if UK had adopted the exact same approach, the volume of breaches would have been indefinite lockdown:

The question of whether the UK could – or should – do this is a key unanswered one when it comes to pandemic preparedness, I think:

Proof, truth, and infectious disease Podcast Episode · The Lancet Voice · 03/07/2025 · 38m

Enjoyed talking proof, truth, and epidemiological research – from policy to public engagement – on the Lancet podcast: podcasts.apple.com/gr/podcast/p...

Unfortunately, it's the quality of reasoning being applied by those loudly claiming that 'COVID measures reduced transmission but had no impact on deaths'.

Even if they claim to have accounted for timing of measures, what matters is measures *relative to epidemic dynamics*, as above example shows.

New Zealand introduced lockdown in the same week as UK in 2020. Over the months that followed, the average stringency of UK measures was higher than NZ, and it had many more deaths.

Therefore, COVID control measures have no effect on deaths...

Which is an absurd claim, right? 🧵

Exams won't tell us whether AI has reached 'peak intelligence' Acing hard questions isn't the same as doing hard thinking

I previously wrote about how acing exams isn't a particularly useful test of AI thinking in the real world: kucharski.substack.com/p/exams-wont...

Won't be trusting an LLM with my tax return any time soon... (from: openai.com/index/introd...)