24-07-25 Roundup

Getting stuff built

Tornagrain is a beautiful new neighbourhood being built near Inverness. A write up of it here


Hawaii has passed Senate Bill 3202 which will allow two accessory dwelling units, or ADUs, to be built on residential properties. Link


Camden Council plans to pedestrianise Camden High Street


The American Elevator Explains Why Housing Costs Have Skyrocketed


Twenty million Londoners: the solution to Britain's housing crisis


Two very good briefings published by UKDayOne:

New Towns for a New Britain: Where Britain’s next new town should be

A Firm Power Strategy for the UK


And three other good posts about nuclear power in the UK published recently:

Why Did Nuclear Flop in Britain?

Revisiting the UK’s nuclear AGR programme: 1. On the uses of White Elephants

Revisiting the UK’s nuclear AGR programme: 2. What led to the AGR decision? On nuclear physics – and nuclear weapons


No Expensive UK Housing Market Builds Very Much Housing

Science / Tech

Will the growing deer prion epidemic spread to humans? (TLDR you're probably alright)


Monumental Proof Settles Geometric Langlands Conjecture


Layer upon layer of unimaginable horrors discussed very objectively in this piece

Raising Welfare for Lab Rodents


Why OpenAI Could Lose 5 Billion This Year

And new Llama and Mistral models got released this week claiming to be close to frontier level. Lots of startups burn money during a land grab. Uber is the iconic example. Lots of journalists at places like the FT now look very silly.

But it's not obvious to me that there are outsized returns to being the no. 1 company in LLMs like there were in ride sharing or search engines or social media. The things that make LLMs better are more flops, data, and labelling. But crucially not the data from user interactions, at least not until there is a new RL breakthrough or we're able to deploy models in an environment with more useful feedback.

It's a great time to be a user with competition this intense. But it does make me think the labs aren't sitting on any major breakthroughs. Otherwise why are their models all so similar? The progress now seems to be more about grinding out incremental improvements to the training pipelines and driving down inference costs.

Miscellaneous

Ben Thompson's recent pieces on the harm the EU is doing through overzealous regulation have been excellent. Brexit was probably net bad but I spend a lot of time feeling very relieved we've left.

The E.U. Goes Too Far

Crashes and Competition


The strange history of osteopathic medicine


Ineos: The petrochemical empire of Britain’s last industrialist


How Rivian Became the Anti-Tesla


However big you think Everest is, it's bigger


Review: How Language Began


Patronage vs Constituent Parties (Or Why Republican Party Leaders Matter More Than Democratic Ones


Andrea Mantegna is one of the big names on the road from the medieval to renaissance world but isn't a household name like Raphael. This is a wonderful piece about him and the renaissance mind more generally.

The Triumphs of Julius Caesar: Mantegna and the Classical Spirit

Tags: Links