iiinova — where your digital self keeps evolving

Joe Hilton @joebhilton.bsky.social · 09/10/2026

Booting up RStudio but the LLM has learned my exact preferences and edited the source code to replace it with a Gameboy Advance era Castlevania game. My productivity plummets and I lose my job shortly after unlocking the double jump.

010

Joe Hilton @joebhilton.bsky.social · 09/10/2026

Not to belabour the point but I think this is a case of poor science journalism. It's a piece about a health economic model without giving any context on why people do health economics, which is absolutely necessary given most people have next to no exposure to the field.

010

Joe Hilton @joebhilton.bsky.social · 09/10/2026

I do think though that results like this are useful in terms of giving you an overall direction for policy; here for instance it's pretty clear that you should try to get as many people as possible working from home before you start thinking about closing schools.

100

Joe Hilton @joebhilton.bsky.social · 09/10/2026

The big issue with optimal control-style approaches is that policies aren't just levers you can pull at will - most of the time it's not obvious what "do policy X at 50% intensity" looks like.

100

Joe Hilton @joebhilton.bsky.social · 09/10/2026

What the KUOW piece totally fails to convey is that this means you could run cheaper measures more intensively while keeping schools open and get the same or better health outcomes. It's about working out how best to use resources given you're willing to take a certain economic hit.

110

Joe Hilton @joebhilton.bsky.social · 09/10/2026

The study simulates interventions identical to the ones pursued in the US but with schools kept open, and estimates that the proportional cost saving is larger than the proportional increase in COVID-19 cases. This suggests that other measures prevented cases for less money per case averted.

100

Joe Hilton @joebhilton.bsky.social · 09/10/2026

This is where the $26 million figure comes in. In isolation this might look like a reasonable cost to save a single life, but it's only meaningful in the context of other measures that could do the same thing. Getting this context is precisely what the authors of the BMC study were trying to do.

100

Joe Hilton @joebhilton.bsky.social · 09/10/2026

One consequence of the cost/benefit framing and the fact that monetary savings can themselves buy further health benefits is that in general you shouldn't choose policies that cost more than something else you could do to get the same health outcomes, since you could just more of that instead.

100

Joe Hilton @joebhilton.bsky.social · 09/10/2026

Importantly, this isn't just about saving money, since the money you save can itself be used to get better health outcomes. If you replace a drug with one that costs half the price, you can give it to twice as many people. This can be a net health benefit even if the cheaper drug is less effective.

100

Joe Hilton @joebhilton.bsky.social · 09/10/2026

There's a basic health economic principle underlying this study, which is that different policies involve different costs and health outcomes and that you should try to choose the policy that gets you the best health outcome for the lowest cost.

100

Joe Hilton @joebhilton.bsky.social · 09/10/2026

The first point to make about the study is that it comes out as broadly pro-NPI; their optimal packages of interventions include a lot of everything except school closure. School closure comes out looking bad from this analysis in part because the other policies come out looking good.

100

Joe Hilton @joebhilton.bsky.social · 09/10/2026

I will say straight away that I'm not a fan of the flippant title of the KUOW piece. The $6000 figure isn't good either - it's the estimated per-capita cost of future learning loss, which seems meaningless to me without a timescale. The $26 million is important though, and I'll address it below.

100

Joe Hilton @joebhilton.bsky.social · 09/10/2026

Thread on this since a) I'm a big Maintenance Phase/IBCK fan and think their science episodes are usually very good, and b) I was a peer reviewer on the BMC Global and Public Health paper (link.springer.com/article/10.1...) at the centre of this article and thought it was a decent piece of work.

100

Joe Hilton @joebhilton.bsky.social · 17/09/2026

This is still a distinct scenario from "little guy in the computer" though. I think the distinction between dispatching programmes that can write new code and are hard-to-impossible to provide oversight for vs creating a new consciousness is actually relevant to what the best response is.

110

Joe Hilton @joebhilton.bsky.social · 15/09/2026

In that case I'd put forward latent class analysis applied to discrete choice experiment responses in health preferences research - knowing there's x% of people with a specific set of preferences helps you tailor interventions to maximise uptake, particularly if you link it to demographic data.

120

Joe Hilton @joebhilton.bsky.social · 15/09/2026

Thought I had a really solid example, then when I looked it up it was a latent class analysis. Many such cases I imagine.

110

Joe Hilton @joebhilton.bsky.social · 12/09/2026

(this is a lie, I'm under 50 and thus do not like Bounties)

010

Joe Hilton @joebhilton.bsky.social · 12/09/2026

Seeing a lot of posts about bug bounties. All I can say is it "bugs" me that they took the mini Bounties out of Celebrations!

110

Reposted by Joe Hilton

Andrew Gelman et al. @statmodeling.bsky.social · 12/09/2026

Annals of hype: Did this discovery in 2012 mark “the discovery of the final laws of nature . . . a discontinuity in human intellectual history, the sharpest that has occurred since the beginning of modern science in the 17th century?” statmodeling.stat.columbia.edu/2026/09/12/d...

statmodeling.stat.columbia.edu

042

Reposted by Joe Hilton

Terence Tao @teorth.bsky.social · 11/09/2026

A group of 25 Fields Medalists, including myself, have made a joint declaration on Math and AI: mathandai.org . We welcome additional signatories. See also this article in the Economist announcing the declaration: www.economist.com/science-and-...

mathandai.org

Declaration — Math and AI

Read the declaration and add your name.

422078937

Reposted by Joe Hilton

explore.org @explore.org · 10/09/2026

Time to champion your favorite Fat Bear! 🐻 Submissions are officially open for the annual Fat Bear Week Campaign Poster Contest! Poster submissions: September 10–16 Voting: September 17–21 Top-voted posters will win prizes. Submit at explore.org/contest

07429

Reposted by Joe Hilton

Jordan P. Skittrall @skittrallj.bsky.social · 10/09/2026

Funded PhD in Cambridge: "Mathematical and computational methods to discover and characterise critical genetic features in RNA viruses, with application to Rift Valley Fever Virus" Group: skittrallgroup.org Ad: bbsrcdtp.lifesci.cam.ac.uk/project-refe... Info: bbsrcdtp.lifesci.cam.ac.uk/how-apply

Goats in the shade of a thatched hut during an investigation into a Rift Valley fever outbreak in Saudi Arabia, 2000. Courtesy CDC/Abbigail Tumpey

153

Reposted by Joe Hilton

Dr Edward Hill @edmhill.bsky.social · 10/09/2026

Seasonal influenza summary from latest @ukhsa.bsky.social surveillance report. #IDSky #EpiSky 🧪 🗓️ Based on data up to week 36 2026 (31 Aug - 06 Sep 2026) 📊 Seasonal influenza circulating at baseline levels in England. Activity increasing. Report 🔗: www.gov.uk/government/s...

gov.uk

National flu and COVID-19 surveillance report: 10 September 2026 (week 37)

143

Reposted by Joe Hilton

Demetri @phdemetri.bsky.social · 08/09/2026

Ah yes, the millenial problems. Among them: * Navier Stokes * Home ownership * That damn moustache tattoo on your index finger

0237

Reposted by Joe Hilton

Thomas House @tah-sci.com · 07/09/2026

New Preprint: "Symptom clusters in Long COVID in the UK: prospective community-based cohort study using unsupervised machine learning" Long COVID, alongside other post-viral conditions, remains a large problem with many open questions. We use a large community dataset to > arxiv.org/abs/2609.05213

arxiv.org

Symptom clusters in Long COVID in the UK: prospective community-based cohort study using unsupervised machine learning

Long COVID is a condition usually defined by persisting symptoms following infection by the SARS-CoV-2 virus beyond the acute phase of infection. The condition has a significant impact on healthcare s...

1145

Reposted by Joe Hilton

Dr Edward Hill @edmhill.bsky.social · 04/09/2026

🚨 PDRA opportunity in avian influenza modelling in cattle. 🏫 Based at University of Edinburgh with @rowlandkao.bsky.social. 👥 Collab w/ Colorado State University, Linköping University, University of Liverpool, University of Warwick. 📅 21-month fixed term post. ⏳ Apply by 23:59 BST, 29 Sep 2026.

jobs.ac.uk

Postdoctoral Research Associate at The University of Edinburgh

An academic position as a Postdoctoral Research Associate is being advertised on jobs.ac.uk. Click now to find more details and explore additional academic job opportunities.

022

Joe Hilton @joebhilton.bsky.social · 04/09/2026

Makes sense, it's supposed to honour research that makes you laugh [at the methodology], and then make you think [about how bad the methodology is].

030

Joe Hilton @joebhilton.bsky.social · 03/09/2026

Guy Is Nicking the Income coefficient

010

Joe Hilton @joebhilton.bsky.social · 03/09/2026

I knew Italian politics was a *bit* chaotic, but 3.8 years setting an 80+ year record is absolutely wild

020

Reposted by Joe Hilton

Matt Henderson @matthen.com · 01/09/2026

Here I draw a circle by hand. In a universe where it is the perfect circle- what does a sine wave look like? What is its value of pi?

620433

Joe Hilton @joebhilton.bsky.social · 01/09/2026

There's a preprinted version of the work I mention on pharmacogenetic testing here: assets-eu.researchsquare.com/files/rs-905...

assets-eu.researchsquare.com

000

Joe Hilton @joebhilton.bsky.social · 01/09/2026

I'm on a podcast! A huge thanks to the teams at @plusmaths.bsky.social and @juniper-network.bsky.social for having me on to talk about working as a mathematician in health economics.

152

Reposted by Joe Hilton

Thomas House @tah-sci.com · 24/08/2026

New Preprint! "A framework for combined epidemiological-genomic inference to improve estimation of household model parameters" with @joebhilton.bsky.social and Golsa Sayyar. We extend the state space of standard household models to combine genomics with epidemiology. arxiv.org/abs/2608.21094

arxiv.org

A framework for combined epidemiological-genomic inference to improve estimation of household model parameters

Models incorporating household structure, with different rates of transmission within and between households, are widely used in infectious disease epidemiology. These models can be calibrated using f...

0164

Joe Hilton @joebhilton.bsky.social · 24/08/2026

Devastated to report that LLMs still lack the sophistication necessary to help me cheat at the Guardian cryptic crossword.

A Google search for the phrase "artist name ending in u" with an AI overview listing musical acts with names beginning with U.

120

Joe Hilton @joebhilton.bsky.social · 24/08/2026

As someone who's spent their career gradually moving from maths to health science, my deepest unresolved maths guy conviction is that you should almost always be writing methods sections in the active voice. You're the one that did the science! Just own up to it!

040

Reposted by Joe Hilton

Nathan Doyle @nathandoyle815.bsky.social · 24/08/2026

Delighted to share that the second paper of my PhD, "The case for using flexible healthcare capacity constraints to optimise pandemic control strategies" has been recently published in Journal of the Royal Society Interface! 1/5 royalsocietypublishing.org/rsif/article...

royalsocietypublishing.org

The case for using flexible healthcare capacity constraints to optimize pandemic control strategies

Abstract. During the COVID-19 pandemic, a key challenge was designing control strategies that balance the benefits and costs of public health measures whil

152

Joe Hilton @joebhilton.bsky.social · 24/08/2026

"Your honour, surely it is self-evident that a nat 20 that lands on the floor is just as much a nat 20 as one that lands on the table"

020

Reposted by Joe Hilton

👹woke 2.0 gremlin👻 @fabius.bsky.social · 20/08/2026

I've honestly really come to hate the term bubble. *Everyone* is in a bubble, it is called there necessarily being a horizon to your personal knowledge and experience! That's just human life!

533722

Joe Hilton @joebhilton.bsky.social · 15/08/2026

That thumbnail image is 100% confirming my assumption that these people fundamentally understand race in Dungeons and Dragons items.

010

Joe Hilton @joebhilton.bsky.social · 14/08/2026

This is how I look when your data availability statement says "on request". Hate to resort to an appeal to emotion but you should not be putting me through this kind of stress.

010

Joe Hilton @joebhilton.bsky.social · 14/08/2026

Would argue though that blue whales are a derived quantity though given we typically measure them in terms of how many cars you can fit inside their different organs.

010

Reposted by Joe Hilton

Tom Cox @dj-acid-reflux.bsky.social · 13/08/2026

Someone told me "Bluesky is just like Twitter." I argued that was untrue on the basis that last time I had a Twitter account most times I got a new follower their bio said "single & looking for fun" & here when I get a new follower it tends to say something like "professor of rare moths".

492316306

Joe Hilton @joebhilton.bsky.social · 13/08/2026

Based on being a student at the time and keeping in touch with people, I would guess that everyone involved in the big 2010s cancel culture stories has gone through quite a bit of political and personal growth since then, and everyone writing articles about it has quite clearly not.

060

Joe Hilton @joebhilton.bsky.social · 13/08/2026

Love the exponential distribution me

000

Joe Hilton @joebhilton.bsky.social · 13/08/2026

Something happening once doesn't mean it will happen again soon, and something happening once also doesn't mean it won't happen again soon.

130

Joe Hilton @joebhilton.bsky.social · 13/08/2026

So it's a random walk about something, but it's not quite the deterministic trajectory.

020

Joe Hilton @joebhilton.bsky.social · 13/08/2026

For something like Lotka-Volterra you can write down ODEs for the mean trajectory which look like the deterministic system plus a few variance terms - the quadratic terms in the deterministic system mean you end up doing expectations of quadratic quantities so you get second order moments.

120

Joe Hilton @joebhilton.bsky.social · 12/08/2026

Seeing a lot of eclipse posting and very little meteor shower posting. Really makes me question the taste of a lot of people I'm following here.

010

Joe Hilton @joebhilton.bsky.social · 11/08/2026

Well if the state and the free market don't provide them then it probably comes under the purview of the guilds. Just do a 15-year apprenticeship and you'll be able to make your own, and if you play your cards right you might even get in with a chance to marry the guildmaster's daughter.

130