Related, Vinad Prasad on cancer and medical ethics. Various interesting things: most cancer drugs don't pass a cost-benefit analysis, lots of cancer drugs get through the FDA with shoddy trials (And probably don't work at all), and this is all driven by the fact that Medicare will pay whatever the drug maker asks for (With social pressure being the only limit!), and likewise insurance companies will insulate people from the costs. Here's a thread featuring myself, Ben Southwood, and Arnold Ventures' Stuart Buck on this. And this about the FDA; Ben has been looking into the topic recently.
Andrew Gelman on the effects of pollution on cognitive ability
If you have read my Longevity FAQ you have probably seen the part about using methylation clocks to track aging. Well we know have proteomic clocks! Turns out that proteins associated to the immune system play a big role. Test set accuracy was around 0.71 to 0.87 (depending on the number of proteins used) using a linear model trained on protein expression as measured by relative fluorescence units.
A new critique of Mariana Mazzucato's The Entrepreneurial State. I was one of the first one writing critiques of the book, you can find them scattered around this blog.
The quaintly named ML outfit Hugging Face has a review of trends in Natural Language Processing
Steps towards designing chips using ML
Imagine taking over an academic supercomputer and mining bitcoin. It happened in Canada
Looking at trends in the startup ecosystem for the 2020s
The case for remote work
Usually innovation is proxied by patents or R&D spending. Here's an attempt to develop a metric based on text analysis, looking at what analysts say is innovative over time.
Tory Bruno, head of the United Launch Alliance, on why they don't think SpaceX's reusable rockets are as cheap as they claim, and why ULA is not pursuing a similar approach.
One claim found online sometimes is that Scandinavian countries do great is because of the Scandinavians themselves (e.g. Scandinavians in the US also outperform), it's not their institutions. David Shor argues that this is just measurement bias .
How Boom Supersonic leverages cloud computing. A remarkable fact is that they only needed 3 wind-tunnel tests to validate their design. Looks like CFD has gotten better in recent years!
The state of the art on lithium-ion battery cathode chemistry
Breaking Eroom's Law. Remember that drugs were getting more expensive per unit of R&D? For 10 years, that has not been true
Reversible computing and how far are we ()
Healthcare spending has an income elasticity <1 in one given area; and that's compatible with an overall income elasticity >1 at a country level, thread.
A review of attention in neuroscience and ML
And a review of transformers in ML
Paper argues that, among other things, the fear of communist revolutions drove reduction in working hours (As concessions to workers to avoid revolutions). I'm initially skeptical of this, it doesn't look like the general picture that I thought was true plus also most economic historians thought (Back in 1995) that it had been economic growth, not trade unions that caused this. Still, a new piece of evidence to have in mind (haven't read yet).
An overview of what it takes to develop a vaccine
Machine Learning is coming to the world of compilers and hardware design.
Mental health conditions by political ideology and race in the US.
Contrary to what I thought, high stakes does not reduce bias (In an experiment in college students in Nairobi)
Boeing 787's software require them to be turned on and off every 51 days for safety reasons
Evidence against dark matter continues to trickle in
Looking at how individual neurons work by trying to fit neural networks to replicate their behaviour