Research shows doctors and their families are less likely to follow guidelines about medicine. Why do the medically well-informed comply less often? https://news.mit.edu/2022/physicians-medicine-guidelines-1215
The Lead-Crime Hypothesis: A Meta-Analysis — It seems that the estimates of the effect of lead on crime are affected by publication bias; accounting for this reduces the effect to 0-36%. https://gla.ac.uk/media/Media_774797_smxx.pdf
2022: A Review of the Year in Neuroscience https://medium.com/the-spike/2022-a-review-of-the-year-in-neuroscience-ae4e0de082e7 [archived version: https://archive.vn/bMwBy]
Study suggests ancient tree rings might not have chronicled solar flares after all: "lack of relationship to the solar cycle means that Miyake events probably aren’t due to a solar flare … these events are lasting much longer than a solar flare would normally … Miyake event of 774 CE seemed to have lasted years" https://cosmosmagazine.com/space/miyake-events-solar-flare-tree-rings/
MicroRNAs are deeply linked to the emergence of the complex octopus brain https://www.science.org/doi/10.1126/sciadv.add9938
Terminal planetary defences: intercepting asteroids just days before impact. https://arxiv.org/abs/2110.07559
“In the early years of fire insurance, insurance company fire brigades seemingly made little distinction between insured and uninsured properties. They were instructed to attend and help put out all fires. The grounds for this policy included the risk of fire-spread between uninsured and insured properties, the advertising value of the firemen and their engines, and charitable acts for those who could not afford insurance.” https://www.tomscott.com/corrections/firemarks/
“How much of the stuff that our genomes do is actually functional? We set out to answer this long debated question by creating a null hypothesis in yeast and human” https://threadreaderapp.com/thread/1605257357994647553.html
"increase in milk production per cow, from 2000 kg/cow in 1944 to 10,000 kg/cow in 2017" https://cabiagbio.biomedcentral.com/articles/10.1186/s43170-022-00080-z
“Neuroscientists should know that we already have the tools to preserve brains fantastically well at the ultrastructural and molecular levels using aldehyde fixation. Cryonics might be BS (it is in practice), but neuroscience damn well knows how to preserve brains today.” https://twitter.com/KennethHayworth/status/1607546480302391296
Existing smartphones will connect with new satellite constellations in 2023 https://spectrum.ieee.org/satellite-cellphone
“It is extremly disturbing that major scientific organizations are forbidding the publication of research that offends some political sensibilities…” https://marginalrevolution.com/marginalrevolution/2022/10/forbidden-questions.html
Fascinating graph:
1. More evidence for the scaling hypothesis.
2. Fine-tuning makes a huge difference.
3. There is no sign that ML progress is slowing down. Expect much more progress over the next few years.
Paper: Large Language Models Encode Clinical Knowledge https://arxiv.org/abs/2212.13138
I fear that humanity is now less prepared to survive a civilization-ending bioterror attack than it was before the COVID-19 pandemic. Especially in the West, the share of people who would refuse to observe a strict lockdown is now too high to survive such an attack with the minimum viable population necessary for a modern economy to keep functioning.
When people start isolating themselves out of fear, it is too late. With a bioterror attack, you need to lock everything down and stop all travel the moment it is detected. What politician would be willing to decide this now?
China actually managed to successfully keep a zero-COVID policy for more than 2 years. If this had been a bioweapon, China would now stand unscathed, while most other countries would be big graveyards.
The discussion of Tao and Baez is great.
Feynman was a master of continuity arguments as well. Well, in his case, it would be a differentiability argument. The most famous example would be Feynman's integral trick: https://www.cantorsparadise.com/richard-feynmans-integral-trick-e7afae85e25c
The idea is that when integrating a function, it's easier sometimes to think about the function as one instance in a more general parameter space of functions. Then you work out the problem and set the parameter value at the end to get the answer you wanted in the first place.
The beauty of Feynman's trick is that there are an infinite number of different parameter spaces that you can embed a function in. And the key is to choose exactly the right one that will simplify the problem.
Like take a simple function like f(x) = e^x. There are multiple ways to view it:
(1) f(x; a) = e^(ax) when a = 1
(2) f(x;a) = a* e^x when a = 1
(3) f(x;a) = e^x + a when a = 0
(4) f(x; a) = e^(x^a) when a = 1
And so on and so forth. For more complicated integrands, there will be many different places that you can embed a parameter if you so wish. But you have to choose wisely.