Paul Skidmore

I'm not sure it promotes understanding, but there's certainly something about shared civic duty that I'm loathe to give up. Justice seems like it shouldn't be a service that is delivered to citizens but not by them. The problem is for too many people it's not a service being delivered at all.

Busted.

3. Which reminds me: the judge was BRILLIANT. I mean, old school in all the ways that are supposed to be bad. But his instructions were what eventually made it possible to reach a decision we were all happy with. Which is another thing that makes me sympathetic to this move.

2. Stubborn and made their minds up before the prosecution's opening statement was over. In this case the judge's directions eventually made clear that the letter of the law and their personal interpretation of justice were aligned. But it was very easy to see how in other cases it might not be.

1. Yes, but I feel like that choice is another place where there is a gap between the republican ideal and the reality. No one knows each other, one guy sticks his hands up, rest of us go along with it, he turns out to be a dud.

2 jurors made their minds up within 2 mins on the grounds that "anyone coming into their house would get what was coming to them".
1 juror kept nodding off during the trial.
1 juror said she didn't mind what we decided but could we hurry up as she needed to get to work.
Everyone hated the foreman.

I served on a jury for an Old Bailey murder trial. I also think we ended up with the right verdict (the defendants were acquitted on the grounds of self-defence) but the way we got there did not fill me with confidence in the system.

You can burn my house, steal my car
Drink my liquor from an old fruit jar
Well, do anything that you want to do
But, uh-uh, honey, don't wear an Elvis wig when you are a judge presiding over cases in Division 4 of the St Charles County Circuit Court.

In particular 1) it over-emphasises the importance of hyper- personalisation 2) it under-emphasises the importance of IRL social contexts, routines, adults and peers.

Can LLMs be personal tutors? Four big challenges

So much wisdom in this short blog from @daisychristo.bsky.social. Ostensibly it's about the potential of LLMs as tutors, but in the process it offers deeper insights about human learning and why edtech so often falls short of its promise to transform it substack.nomoremarking.com/p/llm-tutors...

The quoted text reads: 5. I believe that we often dissipate much of the political goodwill which our aid programs should engender by too much insistence on detail in our agreements with and programs in recipient countries, by slowness in implementing programs and by falsely raising expectations by engaging in too much planning for projects which have little, if any, hope of realization. These aspects of our aid programs should be carefully reviewed in an effort to assure the maximum political impact. In the new programs which we shall be implementing in Africa, we should make every effort to avoid the pitfalls we have encountered in other parts of the world.

Well-known wet liberal <checks notes> Richard Nixon here on why micromanaging aid was counterproductive (from the report of his Africa tour in 1957 here history.state.gov/historicaldo...) h/t @jdevermont.bsky.social

FWIW, I'm sure practices varied, but I checked old emails and my letting agent in Southwark included a reminder about the forthcoming licensing regime in every monthly statement from Oct 2022 onwards, a full year before it went live.

Terrific paper from @lindseymacmillan.bsky.social and colleagues that should (but won't) put to bed the idea that standardized tests are worse for poor kids than teacher assessments. But lots of fresh insights beyond that, including about variation by subject repec-cepeo.ucl.ac.uk/cepeow/cepeo...

(4/4) That's not about policy preferences per se, it's about definition and the arguments you choose to pick to give you that definition.

(3/x) Namely, to find a way of talking to its own coalition that suggests they actually like them, and are listening to them even if they don't always agree or can't always do what they want.

(2/x) Ezra acknowledges it but in the most perfunctory way - he almost seems annoyed - and skips straight past it. Which is a shame, because I think TNC's putting his finger on something that the left often does very badly (especially at the moment) and the right very well.

I found this the most revealing exchange in the (excellent) conversation between Ezra Klein and Ta-Nehisi Coates.

TNC makes this really important (and subtle) point: www.nytimes.com/2025/09/28/o...

As the great man said, "A simple way to take measure of a country is to look at how many want in and how many want out.”

With any technology, the temptation is to argue that this changes everything or this changes nothing, when the reality at any given moment is nearly always somewhere in between. Then suddenly you look back and it's permeated everything. What feels different this time is how quickly that's happening.

But more than any one task, the thing I suddenly noticed yesterday was the breadth of tasks. Other tech revolutions in my lifetime have been similarly wide-ranging. But the transition from landlines to mobile phones to smartphones to smartphones that can do everything a computer can do took 25 years

The important thing about this last category of tasks is that AI is useful to me precisely because they _are_ within my capability, so I can judge its quality. E.g. none of the email drafts it offered up were quite right, but in that moment, to move forward, I needed to be an editor not a writer.

Some were things that the Old Internet couldn't do. Helping me iterate on the right phrasing for a tricky email. Providing advice on the structure of a potential transaction. Tasks that in the past were certainly within my capability to do, but which it can now help me do better and faster.

In some cases it was tasks the Old Internet has (or at least feels like it has) got worse at. Making new book recommendations for my son based on books he's enjoyed before was supposed to be what recommendation algorithms were for, until they were enshittified. ChatGPT nailed it.

Some were things that the Old Internet could do, but it can do better. Finding a flight that satisfies more complex criteria than the comparison sites can handle. Helping me diagnose and fix a problem with my washing machine.

Yesterday was an eye-opening AI day for me. There was no single, mind-blowing "this changes everything" moment. On the contrary, it was precisely that there were a dozen completely unrelated things, spanning the personal and the professional.

Should anyone 😬 be interested in a very readable single volume history of Liberia, I can recommend James Ciment’s “Another America” www.abebooks.co.uk/978080902695...

(11/11) But that doesn't make it more likely to succeed. If anything, it's a red flag that it won't. The "It's Always The Principal Principle" should make us extra cautious about proposing new structures to compensate for the weaknesses of a Principal, rather than to amplify their strengths.

(10/x) Sometimes we focus on structure because changing these other, lower cost things seems too difficult or not something the Principal is interested in or willing to do. A new structure in this scenario *is* actually the easier thing to do.

(9/x) There are lots of lower cost tools to reach for first: the leaders you pick, the teams they hire, the culture they build, the ways of working and coordinating they develop. The results might not look as neat but if they work, who cares? Conceptual neatness is over-rated.

(8/x) Partly this is because they are really hard and really distracting, with tons of the most senior people's time consumed by managing a messy change process, not focusing on the day job. The track record of Machinery of Government changes is particularly dire for this reason.