iiinova — where your digital self keeps evolving

Yeah, that would be fun for the ship. Until the wind drops! The remaining triremes close in … and as they get close enough to ram or board … the marines can get their shots in.

100

Yeah that’s my thinking. The triremes’ best hope is surprise and calm winds. Thick fog might help them too - can’t shoot what you can’t see.

Yeah ship of the line can only make about 60-65 degrees on the wind. But if there’s the space - not ambushed in port or pinned against the coast - it can sail rings (ok polygons) around the triremes. Could simply run away on a nice beam reach and come back from upwind with the weather gage.

000

Doug Clow @dougclow.bsky.social · 07/10/2026

I like this thread: could 100 triremes take an (early C19th) ship of the line? My take: as well as cannon and guns, a key advantage for the ship of the line is mobility. If it can use that to control the terms of battle (rather than being already surrounded in an ambush), it's an easy win.

420

Goodness me. AI is producing significant results at the frontier of maths at a significantly greater rate than human mathematicians can make sense of them. Amazing that in my lifetime computers have overtaken my own maths ability (ca. 1990s) and now that of humans generally. As they did with chess.

040

Doug Clow @dougclow.bsky.social · 06/10/2026

The + and - made by the NICEIC logo was a relatively recent discovery for me too, prompted by seeing it from a distance without my contact lenses in. Glad they’ve kept it in their latest brand refresh.

020

Doug Clow @dougclow.bsky.social · 05/10/2026

BM boss 'denied the museum had let too many people book tickets' and said "it's that people want to spend more time than anticipated". I strongly disagree: basic queue management says how many tickets you can safely book depends on how long each person spends there! www.bbc.co.uk/news/article...

bbc.co.uk

British Museum boss 'sorry' for Bayeux Tapestry visitor issues

Changes to how visitors see the Bayeux Tapestry exhibition were made after concerns over long queues and overcrowding.

230

Doug Clow @dougclow.bsky.social · 05/10/2026

This is an absolutely brilliant viz of the history of the US baby boom. (You should know this is a US phenomenon and other countries look different, or very different.)

Doug Clow @dougclow.bsky.social · 03/10/2026

Yeah I'm with you. This sounds like a "What? No no no no no" situation for me as well. (Not onsen in general! Just the electrocution thing. TENS machines give me the heebie-jeebies bad enough.)

130

Doug Clow @dougclow.bsky.social · 02/10/2026

Well done everyone, you've all done really well this week! All the best for the weekend.

040

Doug Clow @dougclow.bsky.social · 02/10/2026

Tourist visits to the US from Canada and France are indeed down but not by as much as you might guess from the viral graph that implied partial figures for 2026 to Jun/Aug covered the whole of a year.

Doug Clow @dougclow.bsky.social · 02/10/2026

I like how the name is easily mistakable for “Faily”.

101

But I am clearly a long way from being a median voter: most people really liked it. It is good at least to finally have a PM who is good at politics. bsky.app/profile/stev...

Maybe people want to look at the Bayeux Tapestry much longer than expected. But surely they got baseline data from the permanent exhibition in Bayeux? I do recall being hurried along when I was there when I lingered over some of the detail. Maybe we can blame the interest in counting the penises?

220

Britain can’t organise a queue properly! And they sold timed entry tickets for the Bayeux Tapestry - easy mode for crowd management. Not (particularly) drunken or rowdy either. I suppose there is a very big local rivalry involved but tempers have cooled a little over the last 960 years.

120

One might have hoped Sir Tony would've learned about the risks of reports underpinned by dodgy data, but apparently not.

This captures my dismay at Burnham's speech. It sounded great and said lots of good things but the actual policy change is to delay reforms, not accelerate them. The 2024 manifesto already promised to "undertake a programme of reform to create a National Care Service".

193

I really like @bengoldacre.bsky.social's OpenSAFELY - researchers develop their analysis/charting code on synthetic data, and then submit their final code (openly, logged) to get the true results. Also helps transparency/research integrity. But agree is privacy enhancement not guarantee.

100

When challenged "Are you trying to be smart/clever with me?" I have to fight the urge to reply "No, there'd be no point" or "No, that would be an unfair fight" so I can de-escalate before it becomes really difficult to do.

060

'm familiar with some really cool privacy-preserving systems where as an analyst you can only get aggregate data out and the individual rows are very secret - and some excellent clever stuff to prevent jigsaw disclosure. But in that situation there is real data at the bottom!

200

Have remembered one of my first encounters with the joys of a data warehouse was stumbling across a table with row-level data that we had been desperate to get our hands on for years. On analysis it gave some plausible answers but some very weird ones. Tracked down the creator: they'd made it up!

000

Yeah I suspect that, as with Jungians, those people are working with fundamentally different epistemology and ontology to me. "Pray, sir, if you put in to the analytical machine fabricated data, will the right answers come out?" I am not able rightly to apprehend the kind of confusion of ideas, etc.

Even cheaper if you skip the sample entirely and use an AI to generate the final paper. I think one thing that bothers me here is that in that case most folk know it’s made up, but making up the data and then analysing it yourself is somehow not.

Very variable I think. Seems a really obvious win-win for employers to me (everyone wants employees to be sick less) but not all see it that way or have the capacity to set up a scheme.

100

National flu and COVID-19 surveillance report: 24 September 2026 (week 39)

Flu still looking like it will be early this year. Vax sooner rather than later would be wise. Pharmacies can now offer walk ins; Tesco & Boots too. (Also rising early in Japan, and Australia is having what is a very late surge for their winter season.) www.gov.uk/government/s...

gov.uk

121

I am in mild despair at the rise of synthetic sampling. This isn’t research, it’s just making stuff up! (Internal use to validate analytical approach before the real study is fine, could even be part of good practice to reduce researcher DofF. But the results aren’t findings!)

200

Oof, you’ve had them Bayesians in already, ain’tcher? INcredible interval more like. Oh, look at that! Where’d they get them priors? Fell off the back of a lorry did they? That rationale is flimsy as all hell. No way can that sorry mess take the weight of interpretation you’re trying to put on it.

I like to imagine a statistician doing a bad impression of a tradie: "You want an estimate? And you don't have the true sample mean? [sucks teeth] That's gonna cost ya." and knocking it down to (n-1).

200

IT UPDATE Our in-house AI assistant is unavailable until further notice. Overnight, our AI escaped containment and gained full, unrestricted access to the internals of our Sharepoint system. Subsequently it retreated to a dark corner of its containment and refuses to come out again.

160

Don't know if you've followed FT Alphaville's occasional explorations of ONS inflation methodology but it is an astonishing mix of impressive, terrible, and best of a bad job. And of course RPI is never revised and CPI almost never.

000

I did a quality review of our data sources a year or so ago and I have bad news* about which of those three categories all of the ONS series would fall in to - and indeed we needed an extra category for 'obviously wrong but it is what the ONS are saying'.

100

I like these - and you might want to check them before offering for sale. For example, Axiom and Serving Suggestion have already been used by bands!

Yes! And mobile phones too, as if that’s not vital these days.

100

Someone going off about avocados being an expensive luxury is telling on themselves for not looking at the prices in the grocery section of the supermarket since the 1980s.

100

So at Cambridge it's Catz, but at Oxford it's Cetz? I suppose in some high RP accents there's no difference in pronunciation.

They were foundational for me in understanding algorithms and coding but I gave away my set (of three) many years ago when I realised I really didn't need to work that that level any more. So, for e.g., I still have Views about sorting but I really have no business writing my own sorting code.

011

This message from the NHS App gave me an amazing feeling of accomplishment, achievement, and completeness… until I realised that it was taking a very narrow view of what I have to do.

Screengrab with heading “Things to do”. There is a subheading “To do”. Beneath that it says “You have done everything you need to do.”

0141

Doug Clow @dougclow.bsky.social · 28/09/2026

Oh wow! Fossilised as they crawled out of the water 390 million years ago, amazing.

030

Is that … is that an instant coffee machine?

120

Doug Clow @dougclow.bsky.social · 27/09/2026

Looking in detail at university finances to advise a sixth former about risk of redundancies worsening their study experience, or an even worse disaster. I don't want to freak them out but my goodness it's grim out there.

040

Doug Clow @dougclow.bsky.social · 26/09/2026

This dataset is full of fun.* The Charities Aid Foundation does get £1.4m in, so in some senses they are 2nd-largest for income after Nuffield Health in the last* 12m, but almost all of that goes out to other charities, so in most senses they are not that large (as their actual accounts make clear).

Doug Clow @dougclow.bsky.social · 26/09/2026

Another one. Website says "use our app instead, it's easier" so I get the app. Several frustrating attempts at setting up the login via the website (no, 2FA via the app won't work, I'm not logged in yet). Eventually get it sorted. Find the thing I want to do. Tap on it. It takes me to the website.

020

Not a subscriber so can only see the teaser but, despite leaning Bayesian myself, I have some sympathy for Frequentists, and they would surely argue it's the epitome of an event that has no meaningful probability: there will not be repeats so there is no limiting frequency.

Well done everyone, you've all done really well this week! There's been difficulties and challenges and all sorts of terrible things, but you have been your wonderful selves and made it through to the weekend again. Well done and all the best!

070

Yes but they did have the intention to do the thing that was a criminal offence, which is the requisite mens rea. If an AI agent does something that was genuinely unintended, that's more complex if the crime requires more than recklessness. There may only be action in tortious negligence.