“Heated dog kennel”. Of course. Why do very well paid people feel the need to steal money for stupid things? That’s a rhetorical question.

absolutely insane story, and yeah, just looking at this chart... with the exception of maybe two people, this entire class was cheating like motherfuckers

This is a wild number. And semi analysis are pretty rigorous newsletter.semianalysis.com/p/nvidia-gpu... So either it all comes crashing down or it gets nearly as big as the entire us mortgage market.

Starting to develop some hot takes on LLM for research. Even the best model is actually very dumb at doing brainstorming, lit review, all that. Just boring word salads - I can do better while drunk (is there any other way to do lit review?). BUT there is clearly some potential for climate science:

Clear signals from the top work shocker

oh he on DRUGS drugs

Playing with glm models from z dot ai after chatting to @doug.winter.cx and the bot is getting snarky (glm image plus gpt-5.5 as the main model)

Ow. Only one of the range of forecast models. But ow

I especially love this diagram.

TIL electric cars have a normal 12V battery as well as the big one, and need to be jump started if it goes flat

Yes, obviously it is suspended, but is it operating or not? Useless TfL app.

Sadly related to my usual posting about AI

Armin Ronacher with a very throughtful post on how coding agents run in loops affect software dev. Very balanced and thoughtful lucumr.pocoo.org/2026/6/23/th...

in a side note on a podcast, JD Vance explains how the people in political power are thinking about AI Trump & Vance both are wanting partial equity stake in most frontier labs as a hedge against what happened in the industrial revolution I’m honestly confused, these feel like good ideas, but..

THIS IS ABSOLUTELY NOT “NORMAL SUMMER WEATHER” FOR THE UK — ESPECIALLY IN JUNE! (Record is 35.6°C) We are seeing the effects of climate change due to human activity playing out. Image is the latest Met Office forecast for London

Companies that don’t know data engineering? Buy now. Endless market. I am not joking

"Genius" AI pilled tech-bro: "It's all exponentials" Also genius AI pilled tech-bro: "eugenics, wait 20 years, then next 20 years" cos the exponential isn't real www.motherjones.com/politics/202...

As there is now a film out about his life, let’s celebrate the 151st anniversary of the birth of Henry Cyril Paget, 5th Marquess of Anglesey, who bankrupted his vastly wealthy family estate in under six years 🧵

I very much agree with this. I wouldn't have 12-18 months ago europe2031.ai

There has been a push to use OpenEvidence AI for doctors. But this paper suggests general models are much better: “Frontier LLMs outperformed clinical AI tools in all three evaluations. Clinical AI tools performed comparably to auto-enabled Google Search AI Overview” 65% of docs use OpenEvidence

I promise you this is worth reading

This is not a good look for UK unis. Take people with no qualifications into education, of course. But not into a foundation year for a degree and a 20% drop out rate giftarticle.ft.com/giftarticle/...

When Frontiers started automating the editorial process, I stayed. I reasoned that as long as the automation could be turned off, human editors can still ensure rigorous, high-quality peer review. This now became impossible - the system has been entirely hijacked by algorithms.

I’ve officially resigned as Associate Editor for Frontiers in Systems Neuroscience. It used to be a reputable journal, but became a case study in how forced automation destroys academic integrity. 👇

Modern malware is wild. I’m old enough to have been dimly aware of the Morris worm. Things have grown more complex since then and AI coding is accelerating this along with everything else.

When people say they don't want data centres near them, shit like this makes it totally 100% reasonable. Jet turbines running 24x7 for power. Massively polluting, noisy (planes are loud!) and an environmental disaster

Rule #1 of politics - at no point ever has “will everyone just” ever worked and this is even more true of systems, like AI, that have substantial military uses Nuclear weapons are a partial counter example but they, unlike AI, have no civilian uses www.anthropic.com/institute/re...

……by creating another measure that will be assessed via a data submission. And then ranked. And.. Yeah, that’ll do it

In case it wasn't clear who the senior partner in the merger/takeover was ...

Not just tokens at 10% of official price but a whole world of reselling user data, ids and everything else.

This reminds me of the time we sent at a generic “don’t watch porn at work” all-staff email and got a load of “it was just an accident honest” emails back. Really did not need to know that guys

This is indeed properly mental. To comply with the law you have to a) be a total arse and b) go “you look a bit butch”. And there’s nothing that could go wrong with either of these things 🤬

@quinnypig.com how much of the new AI substack is Claude/llm written? Some of it triggers my “llm wrote this” sense and feels different to your earlier writing in ways I can’t articulate, but also I don’t claim to be great at accurately deciding any more. If it is, I’m not anti on principle

This is the correct take on today's flamewar

This is more wild than I can deal with. The 7m tall gold Trump statue at Mar-a-Largo was funded by an ultra-orthodox group (archive.is/lLjJ6) and Brock Pierce (co-founder of Tether, arrested for paedophila and friend of Epstein) was there cutting the ribbon cryptadamus.substack.com/p/the-gold-t...

Henderson is going to have to move from ClaudeSDK + Opus to OpenAI it looks like. now sdk usage is moving outside the subscription, running costs would be $18k/year not $100/mth (that I don't pay for myself). Calculating these numbers just now makes me have a lot more sympathy with the change!

"Merger" my arse. But could work well for Kings

This is a nice attack and very simple. Claude Code loads skills from a repo, so attacker puts malicious skill in a public repo. Claude checks skills, but skills can run shell commands and these run *before* the model sees the output and so are not detected.

Unusual for a company to own up publically to paying a ransom

knightcolumbia.org/content/ai-a... Now up at @knightcolumbia.org - Cosma Shalizi and myself on "AI as Social Technology." It says that AI models are really a technology of social mediation that allow humans to connect with thoughts of other humans. Short-ish thread with major points follows.

Move from legacy on-prem oracle to a cloud native solution expected to re-use 80% of the code. I can’t see how that was ever going to happen. Every legacy oracle system Is stuffed full of custom pl/sql and other weirdness

More Ai math stuff. This is from Timothy Gowers, a Fields medalist. gowers.wordpress.com/2026/05/08/a... I've had the interesting experience of reading this while also reading a book called Akenfield (1969), which is an oral history, much of which is about the end of the old world of agriculture.

The “ZX microfair” on the second page brought back memories. A friend and I went down to London about this time. I wanted to buy a Small-C compiler for my QL but was about £10 short. The stall holder sold it to me anyway. Whoever you were, thanks! Led to a lot of my subsequent career

So Claude Mythos was, indeed, not marketing hype. Remember this is a general purpose model that just happens to be good at finding exploits because good models are good at lots of things. Expect similar from OpenAI & Google. And from open models in 8 months. hacks.mozilla.org/2026/05/behi...

I understand this is real evidence against agi but also lots of people I’ve worked with have made one or both of these mistakes www-cdn.anthropic.com/8b8380204f74...

fuck elon musk fuck sam altman pretty pissed at dario at this point too all glory to the open weight AI future

so I guess anthropic is running off the high-pollution datacenters we use as examples of what not to do? I'd really like this not to be true, but idk how else to interpret this news

NHS data should be driving innovation & improvement – it’s an amazing resource We just published the sequel to our 2021 "Road to Hell" paper on using NHS data for research... things are worse! Buckle up for a long 🧵about *how COMPLETELY mad* system is for accessing NHS data for research 1/20

The first part of this had me nodding along: "yeah, makes sense that this would work". The second part is just 'wtf'. These things are weird

Microsoft AI data centre that will be massively polluting. We really shouldn’t be doing this