iiinova — where your digital self keeps evolving

Tom Smith @analyst42.bsky.social · 19/07/2026

I agree with you. LLMs reward good practice, and punish its absence. If you're doing meaningful work with AI without the backstop of git, how to know what is changing and when? Matt Pocock (YouTube) has helped me develop a sense of what is productive vs potentially uncontrolled change.

021

Tom Smith @analyst42.bsky.social · 19/07/2026

This is a great post. "Not making a decision is still a form of making a decision"

030

Reposted by Tom Smith

Achim Zeileis @zeileis.org · 15/07/2026

⚽ Probabilistic forecasts for today's match at #FIFA2026 In Semifinal: #ENGARG #ENG #ARG Probability to advance: 58.2% vs. 41.8% 🥅 In normal time: Mean goals: 1.4-1.1 🏴󠁧󠁢󠁥󠁮󠁧󠁿 43.9% Draw 26.7% 🇦🇷 29.4% Heatmap with probabilistic forecasts for the possible outcomes of the match in normal time:

Heatmap with probabilistic forecasts for the possible outcomes of the match. The most likely outcome is 1-1 with a probability of 12.7%.

021

Reposted by Tom Smith

Nicola Rennie @nrennie.bsky.social · 10/07/2026

The benefit of making things in #RStats and Quarto is that it only takes about 10 minutes to update an entire article when the next year of data is available. You can now explore this personal project about which baby names were popular in 2025 here: nrennie.rbind.io/data-stories... #DataViz

Screenshot of article showing most popular baby names in England & Wales, Northern Ireland, and Scotland for boys and girls in 2025. They are Olivia, Grace, and Freya for girls, and Muhammad, Noah, and Noah for boys.

1193

Reposted by Tom Smith

Olivier Simard-Casanova @o.simardcasanova.net · 07/07/2026

Belgium annihilating the United States in front of their public is the best way to end this scandal. Corruption can only get you so far, especially when it angers your talented opponents. Well done, Belgium 🇧🇪

121524

Tom Smith @analyst42.bsky.social · 03/07/2026

Fable is doing useful stuff for me. I know I prompted in this direction, but Fable agrees with me - it's very frustrating that NHSE continue to scatter their public data products across different websites, with different UI, naming conventions, and no metadata. Still no APIs on this stuff in 2026!

000

Reposted by Tom Smith

Mark Allen @markallen.io · 03/07/2026

Wrote an update to last year's blog post about how I'm using AI during my cancer journey. markallen.io/standard-ope...

markallen.io

Standard Operating Procedure

Not the type of follow-up post you ever want to write.

1104

Reposted by Tom Smith

Bob Wachter @bobwachter.bsky.social · 03/07/2026

Of the many interviews I've done on AI and healthcare, this one (at the Computer History Museum in Silicon Valley) may be my favorite. The interviewer is the wonderful journalist and author Katie Hafner, who happens to be my wife. www.youtube.com/watch?v=_sZ5...

youtube.com

CHM Live | A Giant Leap: How AI Is Transforming Healthcare

YouTube video by Computer History Museum

171

Tom Smith @analyst42.bsky.social · 02/07/2026

Plus finance of course - that but also needs to converge (largely through workforce capacity). Anyway it makes my head hurt. Fable doesn't have that kind of limitation so looking forward to what it can do before being removed again on 7/7.

000

Tom Smith @analyst42.bsky.social · 02/07/2026

Conceiving and building scratchpads of demand and capacity modelling approaches to support nhs hospital performance and resource planning. Outpatients, inpatients, theatres, diagnostics (15+ types), cancer, ED, ambulance waits, elective, emergency, doctor and nurse workforce all need to converge.

110

Tom Smith @analyst42.bsky.social · 02/07/2026

Now that we have Fable back, I've posed it my most devilishly difficult question to think about while I sleep. It's complicated, interconnected, in places not very well scoped or described (in truth because I don't have a good approach for solving it). Looking forward to seeing progress!

000

Tom Smith @analyst42.bsky.social · 29/06/2026

Claude wrote a UK population growth comparison tool for me using office for national statistics data. It uses webR under the hood so you can run the calcs locally to confirm them. thomuk.github.io/claude-scrat... You can investigate by region(s), sub-ICB(s), or Local Authority(s) #rstats

Population pyramids comparing 2026 to 2036 population projections for the Nottinghamshire NHS Sub-ICB

000

Tom Smith @analyst42.bsky.social · 25/06/2026

I've not watched much of the World Cup, but you can't beat this! youtube.com/watch?v=PjV9...

youtube.com

Monty Python's World Cup Team of Philosophers

YouTube video by Lonnie Goldman

030

Reposted by Tom Smith

Dirk Eddelbuettel @eddelbuettel.com · 21/06/2026

BBC article about Brian Ripley of the #RStats Core team given this week's excellent news concerning 2026 Rousseeuw Price for Statistics recognition for him and the core team. www.bbc.com/news/article...

bbc.com

Prof Brian Ripley on developing R to manage the world's data

Prof Brian Ripley has spent the past three decades working on a computer software that analyses data.

419383

Reposted by Tom Smith

Hiro Ohta @xlix.me · 05/06/2026

This artwork by Karl Jilg, commissioned by the Swedish government, reveals how city streets have excluded human beings. Sometimes, imagination conveys our world closer to reality than reality itself.

Artwork by Karl Jilg. A surreal urban scene shows pedestrians calmly walking along a narrow sidewalk suspended above a vast, deep chasm. A man balances precariously on a wooden plank connecting two sides, as though unaware of the danger below. The surrounding environment appears normal, with traffic lights and storefronts, yet the impossible drop emphasizes a disjointed reality. Everyday life continues despite the looming void.

110021

Tom Smith @analyst42.bsky.social · 19/06/2026

That sounds interesting, what's involved?

100

Reposted by Tom Smith

Frank Harrell @f2harrell.bsky.social · 17/06/2026

Today we celebrate the awarding of the prestigious Rousseeuw prize to the R Core team in recognition of their 30 years of dedicated work on R and in recognition of the massive impact of R on the scientific community rousseeuwprize.org/2026 #RStats

rousseeuwprize.org

The Rousseeuw Prize for Statistics

The Rousseeuw Prize for Statistics is a biennial prize to celebrate outstanding contributions to statistics research.

17917

Reposted by Tom Smith

Jeremy Allen @jeremy-data.bsky.social · 16/06/2026

I still log AI's transformation of work at www.thetreeline.pub/p/ais-transf.... I defined the collection, processing, and writing workflow for Claude. I lightly edit final text. It's a semi-automated curation service for me, so I can keep up. You are welcome to it if you want. #AI #work

thetreeline.pub

AI's transformation of work

A running log

152

Reposted by Tom Smith

Tom Smith @analyst42.bsky.social · 14/06/2026

Deploying R shiny apps to your own cloud server should be simple, so I created Lightbulb.is. Free 30 day trial. #databs #datasky

lightbulb.is

Lightbulb.is

Deploy R Shiny apps to your own cloud server.

141

Tom Smith @analyst42.bsky.social · 14/06/2026

If you have any questions, drop me an email at tom@lightbulb.is. I built this, and would be happy to hear your Qs...

000

Tom Smith @analyst42.bsky.social · 14/06/2026

Why do you need to upload an ssh key? Because the server lightbulb creates is yours. Lightbulb can manage it for you (where it's super convenient to deploy apps with a handful of clicks), or if you change your mind you can ssh in anytime to disconnect from lightbulb.

100

Tom Smith @analyst42.bsky.social · 14/06/2026

How it works: 1) Upload an ssh public key 2) Connect to your DigitalOcean account 3) Provision a server from the dashboard 4) Create a site (with a domain and SSL if you want) 5) Deploy your shiny app by pointing to its GitHub repo

100

Tom Smith @analyst42.bsky.social · 14/06/2026

Deploying R shiny apps to your own cloud server should be simple, so I created Lightbulb.is. Free 30 day trial. #databs #datasky

lightbulb.is

Lightbulb.is

Deploy R Shiny apps to your own cloud server.

141

Reposted by Tom Smith

Alexander Doria @dorialexander.bsky.social · 13/06/2026

Not having any EU competitive labs (plural intended) is about to get very painful.

513415

Reposted by Tom Smith

Achim Zeileis @zeileis.org · 04/06/2026

⚽ Our probabilistic forecasts for the #FIFA2026 #WorldCup are out. 📈 Based on a machine learning algorithm combining data, expert insights, and advanced statistical models. 🤝 Joint work w/ Andy Groll & team, @rmichels.bsky.social, Lars Hvattum, Gunther Schauberger. www.zeileis.org/news/fifa2026/

zeileis.org

086

Tom Smith @analyst42.bsky.social · 11/06/2026

This is right up my street, thanks for building it. Just to let you know that something in the layout is a little funky on mobile

000

Reposted by Tom Smith

Libby Heeren @libbyheeren.bsky.social · 11/06/2026

Did you know you can search for R packages by what they do, as opposed to just their names? Look! 👉🏼 rwarehouse.netlify.app @kylieainslie.bsky.social made The Warehouse for just this purpose. Send it to a new R programmer today ❤️ because it's the resource we all wish we'd had at some point #rstats

Screenshot of a website with "The Warehouse" at the top. The user has typed "add interactivity to a ggplot" into the search bar at the top of the website and results below include plotly and ggiraph

113246

Reposted by Tom Smith

The King's Fund @thekingsfund.bsky.social · 09/06/2026

Bringing down waiting lists is important – but how people experience the wait matters too. New findings suggest that uncertainty and poor communication remain a reality for many patients waiting for NHS care. Our new blog explores what needs to change. 👉 bit.ly/3QaSYLS

131

Reposted by Tom Smith

Lander Analytics @landeranalytics.com · 09/06/2026

Do you need enterprise AI infrastructure for a 120M-parameter model? Maybe not-but the answer is more interesting than you'd think. We break down why Chronos-2, DGX Spark, and a simple FastAPI deployment turned out to be the right combination. bit.ly/4eh0FbC #AI #TimeSeries #Forecasting #DGXSpark

bit.ly

DGX Spark Series (Part 3): When the Wrong-Sized GPU Is the Right Call

What we learned serving Chronos-2 from an R-friendly forecasting pipeline

022

Tom Smith @analyst42.bsky.social · 08/06/2026

Hey Benjamin, is the postgres database available anywhere? The golem refactors are simple (implemented in this fork) github.com/ThomUK/glata..., but to confirm this runs I would need to have a db up. I also need to improve the .Renviron handling which is now on the TODO

010

Tom Smith @analyst42.bsky.social · 06/06/2026

Yes, some of the package setup parts of golem can feel a bit heavy. At it's heart it's shiny modules with a standard entry point, which is exactly what you have.

010

Tom Smith @analyst42.bsky.social · 06/06/2026

If I have that use case right - let me do some thinking about load balancing. 👍

110

Tom Smith @analyst42.bsky.social · 06/06/2026

3. Which leaves your Q about load balancing, and (correct me if wrong) I think I understand the reason for that Q. You have a sophisticated app which sits on a correctly sized server most of the time, but which needs more help when a flurry of concurrent users arrive (say month end, or upload time)?

110

Tom Smith @analyst42.bsky.social · 06/06/2026

2. {glatar} uses shiny modules. With some minor refactoring into the golem structure, lightbulb would deploy it directly as of today. I may experiment with this next week. This would allow you to simplify the current actions (leaving only the quality confirmation on merge to main).

210

Tom Smith @analyst42.bsky.social · 06/06/2026

1. Push to deploy is on my (early) roadmap, and that will enable hands-off redeployments of any app just by pushing or merging to main. The CMD check would integrate into github CI/CD as part of the PR to main. CI/CD should prevent code with failing checks from ever making it to main branch.

120

Tom Smith @analyst42.bsky.social · 06/06/2026

Hi, thanks for this detail. I see how you have things set up - very slick! A couple of comments from me.

120

Tom Smith @analyst42.bsky.social · 05/06/2026

I'm not sure that I follow - the package itself kicks off a container deployment? At the moment lightbulb will handle both conventional app.R entry points, as well as apps packaged up with {golem}. If you can share code I'm happy to investigate what it would take to handle your case.

110

Tom Smith @analyst42.bsky.social · 05/06/2026

I've assumed the people this will interest the most are freelancers, researchers, students, and developers. People with a decent range of apps to host, where the chore of manual deployment is necessary because of the control they need over the server (because of customer, sensitive data, etc)

010

Tom Smith @analyst42.bsky.social · 05/06/2026

Thanks! Honest answer - it doesn't yet. I've tended to operate below that level of complexity and simply size servers based on the app traffic they receive. Open to ideas though - very interested to understand your use-case...

110

Reposted by Tom Smith

Tom Smith @analyst42.bsky.social · 04/06/2026

Sooo #rstats, I built a thing... Deploying R shiny to your own cloud server is still too hard in 2026. Lightbulb.is is my answer - connect to your digital ocean account, click to create a server, and point it at a shiny GitHub repo. That's it - spend the time you save making your customers happy!

lightbulb.is

Lightbulb.is

3349

Tom Smith @analyst42.bsky.social · 05/06/2026

Thanks for the question - I should write about that a bit more. Cheers!

120

Tom Smith @analyst42.bsky.social · 05/06/2026

In fact, after deploying you can completely sever the link between lightbulb and your server if you wish, and continue to maintain it manually.

010

Tom Smith @analyst42.bsky.social · 05/06/2026

Having rented the box, lightbulb believes you should be able to put as many domains and apps on it as you wish, and the user hours uptime is 24x7 on all of them (it's your box). You get full ssh access, lightbulb just helps you manage it.

120

Tom Smith @analyst42.bsky.social · 05/06/2026

Posit cloud is the larger product, but as a user the same limitations as on shinyapps still apply. It's not your server, the number of domains and apps are limited, and user hours are capped. Lightbulb is a deployment convenience tool for a server you also control.

220

Tom Smith @analyst42.bsky.social · 04/06/2026

Lightbulb.is is a tool to make deploying apps to Digital Ocean really simple. You give lightbulb your API key, and from there can do all your deployment from inside lightbulb. No need to worry about nginx, ssl, or the command line! Just clicking to deploy working shiny apps to users...

120

Tom Smith @analyst42.bsky.social · 04/06/2026

Digital Ocean is a cloud server provider. You rent a cloud server and are billed for the time you keep it alive. The smallest costs 6USD per month (less than $0.01 per hour) and is capable of running lots of low-traffic shiny apps. www.digitalocean.com/pricing/drop...

digitalocean.com

Droplet Pricing | DigitalOcean

Helping millions of developers easily build, test, manage, and scale applications of any size - faster than ever before.

100

Tom Smith @analyst42.bsky.social · 04/06/2026

I started building this quite a while ago, and have been using it for my own deployments for over a year. Finally polished up and something to be proud of. Feedback most welcome, either here or by DM. Thanks!

010