John Rogers wanders the City of London’s passages and alleyways

I sometimes wonder if one of the reasons the "London isn't safe" propaganda has gained such a hold is that the city has changed so much in recent decades. I hardly recognise the skyline from the days when I worked there in the 1980s.

But maybe London has always been like that. Go back another 40 years from when I knew it well and you would find a very different cityscape of bombsites and ruins.

Still, a lot of "the old London" remains. In fact, I was surprised how much of it John Rogers found on this walk.

Here's his YouTube blurb.

Join me on a fascinating London walk through the City of London's narrow alleyways, passages and lanes, where many secrets of its past are revealed. 

We explore the rich London history, discovering historical plaques that mark sites like the Worshipful Company of Masons and Jonathan's Coffee House, a pivotal location for early stock market activity. 

This journey into hidden London offers unique facts about the city's enduring legacy from Roman London through Tudor London to the modern day.

John Rogers has a Patreon account to support his videos and blogs at The Lost Byway.

The fault, dear Mandy, is not in your stars, but in yourself

Embed from Getty Images

Someone once  asked Max Clifford: "Max, if you're so brilliant at public relations, why does everyone think you're a cunt?"

Similarly, I would like to ask Peter Mandelson why, if he's such a master of the dark arts, he's always being found out.

Perhaps the answer is to be found in this character sketch by John Crace:

Betrayal is Mandelson’s lifeblood. It’s there in his treatment of Wes Streeting. Poor trusting Wes. A man more used to stabbing others in the back. Wes looked up to Mandy. Treated him as a mentor. How did Peter repay him? By bitching about him being "pathetic" and going through an "early mid-life crisis". 

Then there’s Pat McFadden. Peter encourages Pat to confide in him. Gets him to say the government is directionless. That Keir is weak. That Labour MPs just go on about what taxes to raise so they can give welfare payments to others. Pat’s reward? To be dismissed in an email to Patrick Vallance as an insignificant lightweight.

Nor is Keir Starmer spared. There’s no sense of gratitude for the prime minister having taken a punt on him for the Washington job. For Peter that was no more than he had rightly deserved. The culmination of a lifetime’s brown-nosing the rich, the corrupt and the powerful. So Mandelson happily trash talks Keir to anyone willing to listen. "Rubbish in, rubbish out."

He doesn’t even bother to conceal what he's doing. He’s never happier than when he’s promoting discontent and division. Turns out he hates Labour every bit as much as the Tories do.

People who never use AI are very bad at spotting AI-generated text

AI use is seeping out of business and science writing and into the  world of literature. And that, says Malin Hay in a post on the London Review of Books blog, is a problem, because literary editors may be the worst equipped to spot text generated by it:

Experimenters in the US last year showed nine subjects a series of articles, half written by humans and half generated by ChatGPT, Claude and other large language models. Asked to guess which of the texts were human, the four subjects who rarely or never used ChatGPT in their daily lives scored "at a similar rate to random chance", while the five who used chatbots almost every day at work collectively misidentified only one in three hundred texts.

So how do you spot AI-generated fiction? Hay says familiar tells include a fondness for em dashes and for the formulation "not x, but y", which it has favoured since GPT-3. But at the start of her post Hay quotes three passages that use neither of these but still have a distinct whiff of AI.

She offers suggests some more sophisticated indicators of the use of AI fiction:

Some of the markers seem to be lexical: AIs like talking about sweetness, loudness, quiet, age and beauty. There is a lot of insisting in AI-generated texts, as well as a lot of promising, a lot of permitting and a lot of filling up. 

Another sign is the overuse of tricolons ("something neat, something pleasing, a quiet violence made beautiful"). And bots often leave out definite articles from phrases where they’re not strictly necessary: "Coffee and cocoa leaned wild", "rain in teeth" or, from later in The Serpent, "Sita became obstacle by existing."

There is a flatness or evenness to AI-generated texts: Wikipedia's guide to detecting AI says that LLMs "tend to omit specific, unusual, nuanced facts" and "replace them with more generic, positive descriptions". 

The strange thing about this evenness is that it isn’t usually couched in neutral language: "a flood that chokes but insists upon being swallowed" is violent, but not vivid. Even if you had your face shoved into a cake, it wouldn’t "flood" your mouth. The prosody is so smooth that you feel a lack of pressure despite the description of gross or vile acts; it rings hollow. It fills you with its nothing.

The Serpent is a short short by Jamir Nazir called The Serpent in the Grove which recently won the Commonwealth Short Story Prize, only to give rise to suspicions that AI played a part in its production. You can read it and judge for yourself.

Incidentally, I have long been fond of parenthetic dashes myself, but use en dashes for the job on this blog. That's because em dashes tend to look clumsy in online text. 

This fondness dates back to the days when I wrote for David Boyle at Liberal Democrat News. I noticed that whenever I used brackets in my writing he struck them out, but he let dashes through.