Feedback forms should be active, even linked to github accounts if you have one. These will then be added to github issues logs, processed, and very likely implemented!
28.02.2026 17:40
👍 0
🔁 0
💬 0
📌 0
Feedback forms should be active, even linked to github accounts if you have one. These will then be added to github issues logs, processed, and very likely implemented!
This is broadly how I think about recipes and cooking: at the start there's a list of ingredients; at the end there's a finished dish; and in between there are various processes happening that transform and combine the ingredients in various ways: mashing, frying, baking etc
Another Claudic implementation of an idea I've been playing around with for years:
Flow Kitchen
jonminton.github.io/recipes-as-mus…
(Formerly 'recipes as music')
They’re now ‘an’ opposition party
New @sgsss.bsky.social funded #PhD opportunity to undertake an #ethnography at #galgael in Glasgow with @giopicker.bsky.social and myself. This is a fully funded #Scholarship / #Studentship, which includes a stipend, fees and research expenses. Closes 24th March. www.sgsss.ac.uk/studentship/...
Eg CARs which do some accounting for ‘neighbourliness’ of observations. I think Girosi and King moved in that direction by talking about Bayesian smoothers or similar, and the CBD effectively brings something like this through the predictor terms used
On the broader sentiment of the original post. I think Andy tends to lean into APC as something very flawed but potentially amenable to some fixes. My position’s that it’s often better to start from thinking of the Lexis surface as a real as-if-spatial surface, and using spatial modelling approaches
Similarly Italian data has a pronounced decades waffle like pattern in single year lexis surfaces, likely due to heavy correction of population counts with each census. Again no signals in this 1948 cohort pattern of it being an artefact
We couldn’t find an artefactual explanation. Examples of which include: discontinuities at age 80/90 or so, as this is when the methods in HMD shift and reported raw data often pools into a single age; period only changes, as when a country’s methods change. This was multiple countries, and cohort
And the importance of positive cohort effects in the UK context was argued by Willets in (I think) the 2000s or maybe earlier
On the original point it’s worth noting UK actuaries’ version of Lee-Carter (Cairns-Blake-Dowd) already includes terms for cohort effects. (Not a direct ancestor but close enough)
Thanks, the 1948 signal is intriguing. With this dy/dt display a ‘shock’ is red followed by blue (vertical for period; diagonal for cohort); but with this cohort it’s blue alone. Presence in multiple countries suggests it’s broader than just the NHS
Since its release on Thursday, I’ve been putting Opus 4.6 through its paces. No specific research task, evidence synthesis or review, or even philosophical conversation seems to challenge it, not even remotely.
So instead, I’m vibe coding a game!
jonminton.github.io/boomer/
How I now write my blog (with reMarkable and Claude):
jonminton.github.io/jon-blog/pos...
Very meta. Hopefully specific and detailed enough to show this is a workable pattern for producing online content.
#claude #vscode #Remarkable #writing #cats
The Dilbert Future in retrospect
jonminton.github.io/jon-blog/posts…
#dilbert #ScottAdams #predictions
I asked an Agentic AI to develop a course on how to understand and use Agentic AI. This is what it came up with:
jonminton.github.io/working-with...
It has a core section, then five parallel tracks for different kinds of knowledge worker.
Suggestions and feedback very welcome!
📄 New paper out in Journal of the Royal Society of Medicine with @robaldridge.bsky.social, Rachel Burns, and analysis led by @jonminton.bsky.social
❓How does mortality compare by country/region of birth in England & Wales, 2007–2021?
doi.org/10.1177/0141...
🧵⬇️
New AI cowriting experiment. Are my blog posts better if I don’t write them?!
jonminton.github.io/jon-blog/pos...
I’ve done a cliche! Here’s a Gemini produced infographic of my end of year blog post:
jonminton.github.io/jon-blog/pos...
New blog post. A reflection on 2025, and why it might be the last year we think “nothing’s changed”:
jonminton.github.io/jon-blog/pos...
I thought this was about whale oil!
New experimental stats website (Alpha stage). Interactive tutorials for helping understand GLMs:
jonminton.github.io/glm-dashboar...
Please explore and make use of the bug and feature request options!
New blog post:
Claude Sonnet 4.5 reflects on reading and fact-checking the rest of my blog! (About 200k words!)
jonminton.github.io/jon-blog/pos...
New blog post: WiredClothMother
jonminton.github.io/jon-blog/pos...
I wrote an essay comparing AI to the Maxim gun of the 19th century, then asked four LLMs to produce short ‘rights of reply’:
jonminton.github.io/jon-blog/pos...
Given recent announcements of tougher asylum policies in the UK, I asked Claude AI to track down and summarise evidence on how much of an influence policy is likely to have on asylum claims as compared with other factors:
claude.ai/share/188b49...
(tl;dr: probably not much)
I’m now being overly sycophantic to Claude by suggesting its single letter substitution is actually a really clever shorthand for describing Vince Gilligan’s new TV show #pluribus
As a stats in-joke I thought redesigning the “Breaking Bad” titles to say “Breaking Blind” would be fun, so I asked Claude to have a go at this. It helpfully pointed out that “Bl” isn’t a real element, but decided that “Bk” must be good enough as it’s just 1 letter away.
As a stats in-joke I thought redesigning the “Breaking Bad” titles to say “Breaking Blind” would be, so I asked Claude to have a go at this. It helpfully pointed out that “Bl” isn’t a real element, but decided that “Bk” must be good enough as it’s just 1 letter away.
As a stats in-joke I thought redesigning the “Breaking Bad” titles to say “Breaking Blind” would be, so I asked Claude to have a go at this. It helpfully pointed out that “Bl” isn’t a real element, but decided that “Bk” must be good enough as it’s just 1 letter away.