A speed test result page showing 105Mbps down and 142Mbps up.

Random AP at what is effectively public wifi at the AWS Summit London. I'm trying to pretend I'm not impressed.

And it succeeds.
I've already used it as a definition, in a fraught discussion, in the last hour.
Didn't know how badly we needed that kind of good-provenance baseline until it existed.

That's a really good definition, and a worthwhile scheme.

The problem is the implied adversary: a tech type who games the system and self-publishes a hundred LLM-generated books ~legitimately carrying the logo, just to crow about it.

Trolls gonna troll.

Give me enough data and I can split out the cohort that will never subscribe, so you might as well hit the with a per-article offer.

(It would require systemic, ambient, and inescapable surveillance of every web user, but that's a small price to pay for freedom of choice.)

I will happily campaign for a carve-out in competition regulations if necessary. Anti-competitive behaviour can destroy the economy, but messing with the way I get my brain in gear every morning is the worse evil.

I feel like @nytimes.com and LinkedIn need to have a games coordination council. A numeric CrossClimb is nice, a graphic Connections is interesting. Both on the same day is just discombobulating.

I'm curious to see who gets it right. Big vendors can swallow inference costs, challengers can be more flexible kinda thing. But AI economics are just different.

Collective nouns default to singular, unless you are emphasising individuals within the group.

If that means the world must burn, so be it.

Given that kind of speed, what happens if Iran exploits the massive market leverage that gives it? The opportunities to make money, never mind just sow chaos, are endless.

It's a whole new manifested epistemic security risk that flows from not being able to take the US at its word.

Amazing to see how fast the markets swing when Donald Trump says "good and productive conversations" with Iran – without a breath of counterparty confirmation.

In theory the there is pre-verification risk (of the US just plain lying) built into the *pricing*, but it sure doesn't show as *lag*.

In SaaS, it means customers look for ways to roll their own solutions. In train tickets, the customers set up a grey market.

In both, it means the system is very badly broken.

As it happens, I'm watching the pricing-strategy freak-out in the software-as-a-service (SaaS) space right now. Vendors are selling AI that means customers need fewer users, while the vendors charge by the user. But customers don't like usage-based pricing, which is way more unpredictable.

Advance train tickets tacitly rely on dropouts: the earlier they are bought, the less likely they are to be used, hence the cheaper you can make 'em. It's a close relative to airlines overselling planes.

If the delta gets too big, though, you get the genteel, peer-to-peer equivalent of scalping.

🚄 Commuters bemoan ‘eye-wateringly expensive’ fares for packed Folkestone trains Plus: Car park's ANPR cameras to be scrapped after unsuccessful trial, and lorry driver charged with dangerous driving after crash

When users revolt against your pricing strategy:

"... more than 300 regular travellers trade tickets for trains between Folkestone and London, often ones purchased well in advance—to secure the very cheapest fares—but which are no longer needed."

folkestonedispatch.substack.com/p/commuters-...

Bless you, Which magazine and @harryrose.bsky.social, for answering the call and going hard on hot cross buns.

Clearly 'tis the weekend the major bun buying begins.

I spend a lot of time in rooms where deeply serious people have deeply serious conversations about the future of serious journalism.
Which is important, absolutely.
But popular journalism is important too.
Don't underestimate the joy a review of hot cross buns can spark.

It takes guts to lead you (news-site) newsletter with a hot cross bun taste test. And editorial leadership more open-minded than the average.

But if you follow the data, that is what some audiences want. They won't say it when asked, but their clicks speak for them, eloquently.

OpenAI teases new model with "chaotic" First Proof test After "chaotic sprint" attempts with rule-breaking human prompting, OpenAI believes it can now beat First Proof type challenges with a simple prompting strategy.

If you are watching for signs of AI recursive self-improvement, take note of OpenAI's latest claim on #FirstProof.
It's been messy, and the proofs are still not validated, but it points to genuine novel discovery, potentially fully automated.

www.thestack.technology/openai-tease...

We always knew AI would take over. We never knew it would be lobster themed. #OpenClaw

Anthropic and the Government of Rwanda sign MOU for AI in health and education Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.

First government-level partnership deal in Africa for @anthropic.com it is is with...

Rwanda.

South Africa and Nigeria have a lot to learn from Rwanda's approach. It should be at a reputation disadvantage, but somehow it keeps getting the Western outreach.

www.anthropic.com/news/anthrop...

The Sideprocalypse - Johan Halse You can't open a feed today without having AI boosters fling word salad like "agentic engineering" or "openclaw" into your beautiful but disapproving face. I'm terrible at ...

Okay, so the dark side of this from the SaaS perspective, from @johan.hal.se.web.brid.gy today:

"Listen: every idea you've ever had, every single one, some cocaine-addled sales critter has had too. And they're better than you at SEO."

johan.hal.se/wrote/2026/0...

The OpenClaw news today reminds me of this ancient* post by
@commaok.xyz on building an app while shopping. The sub-head has haunted me ever since: "Creating is faster than searching".

The realisation hasn't really broken beyond devs yet, but any day now

commaok.xyz/ai/just-in-t...

* 2-month-old

I love ads. Not only because they have largely paid my bills all my life, also at a philosophical level.

But those damn Anthropic ads about AI and ads are living in my head. I may have to switch from ChatGPT to Claude for language work just to avoid thinking about the big picture a little longer.

In the late 90s, the CIA World #Factbook was indispensable. There weren't many comprehensive reference works freely available, and it came with rock-solid provenance.

Now the source is untrustworthy, with the US going post-truth, meh, no big loss.

www.cia.gov/stories/stor...

Phillip de Wet | How do you solve a problem like democracy? | News24 UK liberals are trying to pre-empt Reform UK taking control of the BBC, and find themselves facing a fundamental problem, writes Phillip de Wet.

This is why writing columns rules.

"Nobody was in an armchair, nobody was smoking a pipe, and I was probably the person closest to having a beard, but it works better if you imagine everyone sitting around in armchairs, smoking pipes, and stroking their beards."

www.news24.com/opinions/col...

Wow. Talk about real economic evil. The amount of value destroyed – and the tiny relative size of the gain for the wrongdoer – is mind-blowing.

Mill Media journalist faces £10,000 county court bill after exposé Mill Media is being sued for libel by an Italian businessman over reporting about his alleged activity in London's rental market.

But intimidation of some sort seems to be a common thread. I only know about the @londonermag.bsky.social piece because the subject is going for legal intimidation.

pressgazette.co.uk/publishers/d...

'Claudio is Scamming' How an Italian charmer is duping the owners of London’s priciest properties

The first world is so weird.
In Johannesburg, buildings are hijacked on an enterprise basis, with (non-owner) slumlords using violence and threats as a profit driver.
London's equivalent is "rental arbitrageurs", who unlawfully sub-let and don't pay the rent.

www.the-londoner.co.uk/claudio-is-s...

Gaining traction in Japan this week: some eye-popping results for LLMs on the Common Test for University Admissions.

Lifeprompt has been running the experiment for four years. Gemini and Claude did pretty well, but look at GPT! Red means a perfect score.

lifeprompt.net/news/w-j19fvl

In the year 2026, a pretty serious organisation – with support staff and consultants and a tall leadership pyramid – took the better part of 3 working days to realise the statement "We do not use AI" is false.

For the sake of clarity, if they touch a computer, your people use AI.