What am I up to now?
July, 2026
Contents
Updates
The distinction between Google and Google DeepMind is real, but it is not quite the distinction between two companies. It is more like the distinction between a large state and an unusually autonomous city within it: shared resources, shared interests, and a stubborn local identity. This dates back to Google’s purchase of DeepMind in 2014:
Google had allowed [Demis] to retain DeepMind’s independent culture in London. Google had even granted his followers a privileged status: DeepMinders could get into any Google building globally at the swipe of a key pass, and help themselves to the free food; but Googlers were barred from DeepMind’s premises. (Mallaby 2026, p.162).
There are a few differences between Google and Google DeepMind offices, but the most consistent is the type of bacon served in the office cafeteria. All of the DeepMind cafeterias I’ve visited serve back bacon, what one would traditionally find in a full English breakfast, while the Google cafeterias serve streaky bacon, the most American of cuts.
After three weeks at DeepMind, I love the smart people and independent atmosphere on my research team. For a good sense of what’s important to our team now, the team lead, Iason Gabriel, was profiled by The Guardian last week. I’m very lucky to have a lot of flexibility to work on what I’m interested in: the effects of transformative AI on lower and middle income countries.
It’s hard to think and write about the economics of AGI because AGI can break a lot of economics. The assumptions fundamental to many of our models aren’t a good fit for the speed of AI agents or the scalability of compute.
The most productive approach is to focus on mechanisms: we may not know how the world looks after AGI arrives, but we can assess which levers the technology pushes. Last month, at a conference, I saw two presentations about how AI will affect the size of future firms. In the first, Seb Krier talked about how AI agents speed up bargaining and contracting between specialists, making it easier for each specialist to operate as their own firm. In the second, Erik Brynjolfsson presented his work with Zoe Hitzig on AI and central planning: improvements in accessing and processing data makes centralized planning more feasible.
Both of these are valid and useful economic mechanisms, but they make opposite predictions about firm size. Much of studying AGI’s economic effects is like this. There are viable stories about why AGI might push interest rates up, or down. The debate over Jevons paradox is like this. At DeepMind, I’m going mechanism hunting. Over the rest of the year, I hope to identify the most important ways AI will affect welfare in poor countries, and maybe figure out what to do about it.
I’ll be in London through the rest of July, and back in San Francisco in August. As always, I’d love to chat with anyone interested!
Reading
Elleman’s Modern Chinese Warfare is an overview of eighteen conflicts fought by the Chinese in the 19th and 20th centuries. The best parts are really good: Elleman introduces each chapter with a very brief summary of the conflict, and uses each conflict to draw conclusions, some narrow, some broad, on how China has used war in modern times.
This is not, however, a detailed military history, but a history of how China and its enemies use coercive diplomacy. One chapter, on the Xinjiang arm of the Muslim rebellion of the 1860s, breezes over the entire military conflict which caused roughly 50,000 deaths to focus instead on the aftermath’s maneuvering between the Russian Empire and Qing diplomats. And this is a lightly bizarre diplomatic world. The territory at issue, the Ili valley, was contiguous to both the Russian and Chinese empires, but thousands of miles from either capital. When the first terms were revealed in Beijing and deemed unacceptable, the Qing negotiator in St Petersburg reportedly answered: “The treaty having been clearly settled, it is absolutely impossible to re-negotiate.” He was, of course, wrong, and was executed upon his return to Beijing.
Elleman also had a novel take on the Sino-Vietnamese War. In 1979, China invaded Vietnam, and ended the war with a unilateral ceasefire after 27 days. The narrative I’ve always told is that this was a serious Chinese defeat.In those 27 days, about 25,000 Chinese soldiers were killed. In the 19-year-long Vietnam war, 58,000 American soldiers were killed. The Chinese lost 100x more soldiers per day than the Americans! Elleman, however, puts the war in context with the Indian border dispute in the 60s and the limited territorial war with the USSR in 1969. Throughout the 1970s, and after these conflicts, China perceived an increasing risk from the Soviets, and by 1979 was willing to accept a bad campaign to reduce this risk. Elleman claims that the goal of the invasion of Vietnam was to “discredit Soviet assurances of military support to Vietnam”.
This is a novel claim! The Chinese were understandably afraid of a two-front war, and the Soviet mutual-defense treaty with Vietnam make that increasingly likely. In fact, the Sino-Vietnam War seemed to be purely for a Soviet audience. Elleman points out that most of China’s best units were moved to the northern border directly before the invasion. China was perfectly happy to gain no territory and lost tens of thousands of troops, to prove that Soviet defense guarantees had no teeth.
There’s another, almost fantastical element to the war. Elleman links the conflict to the 29th “anniversary of the signing of the 1950 Sino-Soviet Treaty of Friendship, Alliance, and Mutual Assistance on 14 February 1979.” This treaty had a thirty-year expiry, and Beijing was certainly uninterested in renewing it. Starting a war with a Soviet ally on the anniversary of the signing of a peace treaty with the USSR feels, to me, very mandarin. The anniversary timing, minimal and cheap force allocation, and willingness to spend lives to alter other states’ beliefs make the Sino-Vietnamese war look less like the well-understood war for territory than a very loud diplomatic signal.
The book is not particularly relevant to Chinese warfare in the near future. There is practically no naval presence after the Sino-French war, nor air power. But the book does explore China’s use of war-as-communication clearly, and this is a useful frame to keep in mind when interpreting their military actions.
My most anticipated book of 2026 is Zach Weinersmith’s Sawyer Lee and the Quest to Just Stay Home. Yes, it is aimed at ages 8-12, but his Bea Wolf was aimed at five-year-olds and that was one of my favorite books of 2024. Sawyer Lee is Weinersmith doing Wodehouse-for-kids, and I am a huge Wodehouse fan. I can’t imagine this being a flop. Sawyer Lee was released in the US in late June, but it isn’t out in the UK yet. My copy is waiting for me back in San Francisco.
I see why Weinersmith is also obsessed with Wodehouse. His explanation is that Wodehouse’s books have the most “complex and intricate plot work” in fiction. On the level of individual sentences, there is no one funnier in 20th century prose.
There are only two other authors whose farces I find as delightful and polite as Wodehouse. Connie Willis doing comedy is up there, most notably in To Say Nothing of the Dog. But Eva Ibbotson can actually hold her own against Wodehouse. In June, I read a collection of her short stories, A Glove Shop in Vienna. Ibbotson is a foil to Wodehouse. Her plots are rarely as complicated, and she’s perfectly willing to use a deus ex machina to resolve a love story or misunderstanding, whereas Wodehouse’s only deus is the butler.
Ibbotson’s protagonists are young women, universally, they’re cheerful central European refugees, often Jewish, living in England around the World Wars. This is, of course, her own life. She was born in Vienna in 1925, moved to London when she was ten, and married a professor at Cambridge University. She is deeply in touch with cruel history, and this history is often the inciting event in her stories and novels. One heroine is a Russian countess fleeing the Bolsheviks; another was sent to London by her parents after Kristallnacht.
The rest of her stories are rarely cruel, and the rare cruelty is often to a purpose.There is an absurd sequence mocking eugenicists at the end of A Countess Below Stairs which is strictly cruel. But one should make allowances. The horror of the wars and refugees don’t exhaust the supply of weird, selfish, erotic, generous people. Her stories are about women finding homes for themselves among these people. For her refugees it’s a mixed ending. Some manage to recover part of their lost world in England; others assimilate into their new world. But most create for themselves a better and more mixed little civilization.
Links
The reviews of the Obama Presidential Center have been quite negative, although I hear the food is good. It seems like a big miss, especially the architecture.
Here’s a lovely essay about river tubing in Louisiana. It has beautiful imagery.Someone asked if this could have been AI written. It’s from 2003, so I am confident it’s not. But I plan to continue to share things regardless of AI-provenance, as long as I’m sharing them for the beauty of the writing.
You try to keep your distance from the convoys of high-school tubers, who tend to float in circular formations, like threatened wagon trains, around stashes of illegal beer. Occasionally you wave, with veiled condescension, to a fleet of passing canoers, trapped in their aluminum hotboxes and actively assaulting the river with oars.
Sandy Magnus , an astronaut who was kind to me often when I was in the space industry, has a wonderful travel blog. Last month she did my favorite hike, the Hadrian’s Wall path across northern England. What a lovely place.
Relatedly , did we ever get an explanation of the Sycamore Gap vandalism? Apparently not. But the two guys who cut down the famous tree were sentenced to 4 years in prison. They were both in their 30s, not kids as I had expected. Apparently there’s a video.
Is it possible that I was born at the perfect time to experience a radical abundance in floating balloons at birthday parties? The US government managed a National Helium Reserve, starting in 1925, for blimps and, later, rocket coolant.
In 1996, Congress ordered the privatization of the helium industry, and the Reserve sold down its assets below market price from 2005 to 2025.
Professor Richardson also believes that party balloons filled with helium are too cheap, and they should really cost about $100 (£75) to reflect the precious nature of the gas they contain.
This essay about the emerging genre of “programmer science fiction” was a bit scary because of how well it describes the box which fits my interests: Wodehouse is there, as is Chesterton, Lewis, Ada Palmer, Ursula LeGuin; Borges features prominently, as do Susannah Clarke and Neal Stephenson.
And this short story the author recommended was, therefore and predictably, very good.
Finally , two articles about travel. Travel is more accessible than ever, and over-tourism has become a political issue in many corners. One response is some destinations wishing to be removed from the UNESCO World Heritage list.
Despite the falling cost of travel, some places are still neglected. One is West Bengal (once the “richest place in the world”), where I spent a lovely summer in 2017. Kolkata is beautiful and fascinating and rarely visited despite its size and historical importance. Beyond Kolkata, I never got to visit Murshidabad, and only two days in Chandannagar.
Previously
…