Arthur Clune

I was so surprised at how little the government had planned for actually governing

And, random aside, I couldn't find that article using search on the Guardian's website because I didn't hit the exact right keywords. ChatGPT found it in two queries - first got near, I refined and the second was spot on

Optimism of the will Evaluating Labour's plan for the NHS

This post by @samfr.bsky.social is excellent on it

samf.substack.com/p/optimism-o...

Also, for a more positive take, this from the Guardian, which we are not learning from it seems

www.theguardian.com/society/2025...

The overreach in the other direction gets under way...

Sigh.

Wild thread about the Titan sub that imploded

gpt-5 in cursor does some odd things that neither the default model in Cursor or Claude Code don't. It's very keen on running commands via zsh
/bin/zsh -lc 'uv run pyrefly check | cat'

The default model correctly calls
uv run pyrefly check
directly

Update - I do have API access. Curl works fine. Just some of the tooling I've been playing with doesn't recognise it yet

gpt-5 is very, very good at code based on some quick testing. I don't seem to have API access in the UK but access via Cursor works.

It's just a shame Cursor isn't as nice as Claude Code, but it is a lot better than last time I looked at it

How do say many obviously wrong charts make it into usch a high profile keynote?

A chart screen captured from the opanai livestream. It's wrong.

The chart crimes will continue until morale improves #gpt5

A do-or-die moment for the scientific enterprise Reflecting on our paper “The entities enabling scientific fraud at scale are large, resilient, and growing rapidly”

Research on the scale of academic fraud/paper mills. It's not a cheery read. The problem is large, systemic and growing

He had almost determined that, in spite of his promise, he would not go to Berkeley Square before the dinner. But Mrs. Houghton was not of the same mind. A promise on such a subject was a sacred thing, and therefore she wrote the following note to Lord George at his club. The secrecy which some correspondence requires certainly tends to make a club a convenient arrangement. "Why don't you come as you said you would? A." In olden times, fifteen or twenty years ago, when telegraph wires were still young, and messages were confined to diplomatic secrets, horse-racing, and the rise and fall of stocks, lovers used to indulge in rapturous expressions which would run over pages; but the pith and strength of laconic diction has now been taught to us by the self-sacrificing patriotism of the Post Office. We have all felt the vigour of telegrammatic expression, and, even when we do not trust the wire, we employ the force of wiry language.

"Wilt thou be mine?—M. N.," is now the ordinary form of an offer of marriage by post; and the answer seldom goes beyond "Ever thine—P. Q." Adelaide Houghton's love-letter was very short, but it was short from judgment and with a settled purpose.

In 1875 the narrator of the relatively obscure Anthony Trollope novel “Is He Popenjoy?” was complaining about how texting had ruined sex

On reflection, the difference is that in 2023 people were still feeling confident about responses like “it can’t cite sources, so this won’t affect my class.”

As that confidence has faded, we’re seeing a subset of faculty coalesce around a strict zero tolerance position.

I still think we’re getting beyond denial—but wow, skimming the QTs on this you would not know it.

This reads to me as maximally cautious guidance. About as negative about AI as you can get without saying “zero tolerance for any use at all.”

It’s getting flamed for not being zero tolerance.

Which isn’t going to happen any more than a return to block grants 🤷‍♂️

My point is only that we should fund research properly and not cross subsidise

What the latest HESA data tells us about university finances The headlines are well known, but the full HESA release adds a lot of detail to a generally bleak picture. David Kernohan has the charts

Yes. Which creates an existence proof that it’s possible to teach for the fee. And it’s not necessarily humanities that go - History is cheap to teach.

Theres a lot of data in wonkhe.com/blogs/what-t...

The unis in trouble aren’t always the ones I expected. Look at Durham

In the UK it’s a pretty good approximation tho. And RG overall base covers more than just teaching (and they generally aren’t the most efficient)

I agree, but I don’t think a model where student fees subsidise it is a sustainable one.

Indeed. But I believe there are some providers who do cover their costs with fees (large, post-92s with no research) which would indicate that it's possible if you're efficient. What I was trying to say, badly, is that the whole funding system is borked

I'm reviewing SAP contractor CVs. There's a lot of "15 years+ of ECC to S4/HANA migration experience".

S4/HANA came out in 2015

Someone is going to have to tackle the core problem at some point. We have some of the highest fees in the world, and it cross-subsides research

Anthropic cut OpenAI out from Claude Code 😂

It’s many things, but I wouldn’t say it’s shocking.

I really hope they roll that back!

Is this on a progressive rollout? For me (Google Workspace), it still works as it always has

Tweet from Demis Hassabis stating the Google processed a quadrillion tokens in June 2025

Numbers from Google on tokens. Doubling every month. 1,000tn in June.

Incredibly instructive reporting on what appears well on its way to becoming the biggest, most concentrated, most energy- and resource-intensive, and most speculative wave of capital investment ever.
ig.ft.com/ai-data-cent...

It's almost like there could be an other reason why they want to make the argument!

wtf?! It never, ever occured to me that wikipedia might get caught up in this