Unsurprisingly, human capital matters more than buildings for scientific output. Also, some evidence for the Newton hypothesis.
Equipment Supply Shocks. And how Steve Jobs got a law changed to be able to get every school in California a computer.
Inside Fast's (An apparel company larping as a payments company) rapid collapse
Using ML to design hardware for specific NN architecture. Interesting as well how the authors paid attention to economics (considering engineering salaries, semiconductor manufacturing costs, etc)
A case study in early-stage startup execution
Why is AI hard and physics simple?
Riva on toxoplasmosis and the role of germs in chronic diseases. After the Epstein-Barr & MS paper, I think it's worth looking more into this.
Black-White disparities in life outcomes in the US start early in life (ht/ Scott Alexander)
Some new Gwern essays, on teaching reading of scientific research
The Terra blockchain (Anchor, Luna, UST) imploded. Though this is more Ponzi-like than the article says. Proper fractional reserva banking requires more stable assets (short term ones, not mortgages or long term bonds as is common practice now; also FDIC bad, we can have our free market banks and eat our stability cake too)
Ben Reinhardt on housing researchy ideas in startups or not
Gato, DeepMind's new model, at first I thought it's probably the closest to AGI I've seen so far, a single system that can achieve ok performance across tasks as diverse as controlling a robot arm and answering questions. Sure it still makes silly mistakes ("What's the capital of France -> Marseille") but as we know with AI, if the proof of concept kind of works, taking that to working at human or superhuman performance is a matter of more dakka. Some Gato discussion and skepticism in this thread. Also see here and here to curb your enthusiasm.
How does Tenstorrent work?
Scott Alexander reviews a book on Xi Jinping
Andy Matuschak, interviewed
Microbes have structures in them that trap gas, enabling them to float
Aubrey de Grey & Charles Brenner debate
Lab mice are not getting chonkier
A Longevity intervention combination I wanted to see for a while: senolytics and reprogramming
Roger's Bacon extremely spicy essay on progress studies
I ask Twitter "Has anyone used LLMs for science?"
Michael Nielsen asks twitter what to read to understand Transformers
Students are asked to do some data analysis in a dataset where the datapoints are arranged in the shape of a gorilla. A number of them missed it.
Richard Ngo reviews a book on NY nightlife
Ben Kuhn on 1:1s
Book review of Making it in Real Estate
Sarah Constantin's newsletter, lately covering energy storage
Lada Nuzhna on deep learning for biochemistry
Transformers for software engineers
Brief interview with Luke Gilbert, a new core investigator at Arc Institute
Everything Everywhere All At Once was good