Paper argues that the performance (citations) of a scientific team is highly impacted by the performance (citations) of the weakest performing team member.

Preventing the collapse of Civilization, by Jon Blow. He argues software is becoming worse, and that this is masked by ever improving hardware.

A Case for Oxidation: Why the Rust programming language is great

How does the sensation of touch arise, at a mollecular level?

Fujitsu simulates a quantum computer with a classical computer (They built a digital annealer), gets massive speedup.

The largest chip ever has just been built. There is certainly no chip area stagnation!

Michael Huemer on the plausibility of conspiracy theories

Alex Tabarrok comments on a graph that got briefly famous on twitter, showing a disproportionate growth in the number of healthcare administrators (vs physicians).

Andrew Gelman criticises Cass Sunstain's intellectual rigour and attitude towards the replication crisis.

An example of using ML to facilitate the discovery of how to actually synthesise a given chemical species.

Scott Alexander reviews Secular Cycles

Sabine Hossenfelder reviews The Secret Life of Science

Stratechery on Privacy Fundamentalism

A review of scale effects in growth models

Why intellectual conformity is actually good, as a general policy. Also, Huemer on heuristics about when to defer to experts and when to challenge them. And Huemer against tenure.

I've heard a couple of time that there is increased monopoly powers in the airline industry (See for example Azar et al. 2018). Be that as it may, airfares keep going down

From YC, how biotech startup funding will change in the next 10 years. The interesting bits hare are links to companies that facilitate actually doing experiments.

Matt Clancy's research digests, summaries of economics papers

Memory is generally thought to be encoded in synapses in the brain (At Nintil I've linked in the past some papers against that). The latest in this debate is one that claims to have done memory transfer across snails by merely injecting RNA into neurons.

Stupid solutions to real problems in science, by the 100% Confidence Interval blog team

  • Featuring "the Octogon": Borrowing from the Ultimate Fighting Championship, the Cage of Adversarial Collaboration is a ring proponents of competing theories can only leave once they come up with a decisive, preregistered test, the result of which they vow to accept (the interocular trauma test is among the approved finishing moves). Following Max Planck, science advances one funeral at a time – this might speed things up.

    Chime in with your problems and solutions in the comments. We might have enough material for a second post. Maybe next time some of the solutions might actually be real?

Progress Studies manifestos are popping up faster than one can read them!

Research evaluation: In some countries there are periodic reviews of academic output from universities, and this is used to guide budget allocations. In other countries, like Italy, an algorithm is used based on the impact factor of the journal relative to that of other journals in the area, and the relative citations the paper has received. Turns out that there is a high (R^2=0.9997!!) correlation between the budget allocation chosen by the algorithm vs that chosen by experts in the REF2014 assessment in the UK .

Saloni reviews The Gendered Brain

From 1990 onwards, the war on cancer is finally being won

"Open and replicable science cannot save us from academia", and p-hacking has been with us all along, for centuries because those are the incentives scientists face.

No. Academia didn’t change and science wasn’t saved. 200 years after Babbage’s book, academia remains the same, but worse: insecure employments; unhealthy hierarchies; unhealthy work-life balance; unwanted relocation to secure a position; administrative focus on quantity rather than quality; career development rather than scientific development; fear of sharing ideas, data and materials with colleagues; publish or perish. These are the things that make us employ questionable research practices. Academia might once have been created for the sake of science, but if so, that purpose was lost a very long time ago

Decentralized research institutes for researchers outside of Academia, Ronin Institute and IGDORE

I asked on twitter what the most efficient language is for information transmission. The one paper I was aware of that looks at this, which Gwern cites here finds that most languages are similar in information transmission, with the exception of Japanese, being worse.

Higher IQ (weakly) predicts libertarianism, a finding that has been replicated in the US, Brazil, and China, now also true in Denmark.

Apple invests in R&D as much as Spain (private+public ). I wonder in what exactly Apple spends that money.

In my post Fixing Science (Part 1) wondered how many problems are there in the natural sciences, after a brief examination, fake data seemed to be one of the common issues. Here's an example of that in the wild.

Alexey Guzey's How Life Sciences Actually Work: Findings of a Year-Long Investigation

The state of biotech VC

Monads are back! (No, not the functional programming ones, the OG monadas)

Religion correlates with self-reported prosociality, but not when objectively measured.

Nobel Prize winners are more creative (playing music, creating art, performing) than the general population.

In line with my previous link's post "coffee is fake news" bit, drinking coffee (or tea) before bed does not affect quality of sleep. (Alcohol and nicotine did)

DARPA on the history of DARPA

A toolkit of policies to promote innovation

Recent work on new phases of matter

The Science of Science Funding Initiative

Predicting business success is hard (in Nigeria): It's hard, and what works marginally better, relative to the business plan, is knowing the business sector, gender, age, and ability of the entrepreneur. Most R^2s are below 0.2.

However, in the US, for venture-backed firms, a panel of entrepreneurs, investors, and executives is able to see what will be successful and what won't, but only in "hard tech".

Building infrastructure: The Nordics seem to be good at it

How useful are polygenic risk scores for clinical purposes?

Joseph Heath reviews Stubborn Attachments

Lessons from the EA economic miracle. I haven't read much about industrial policy, but my baseline view is that it may have worked in the past, without implying that it can work now, less likely so in developed countries. Deserves more examination. Followup from the author

RCT in Mexico adds to the evidence that management practices matter, and that management consultants can help.

US standard of living has improved more than it would seem from GDP due to improper inflation adjusement.

The Human Brain Project, not a very galactic brain

Simon DeDeo attempts to point out a weakness in determining genetic causes of behavioural traits, fails. Why?

  • First, I don't know what this is supposed to be criticising, and if it's just a snarky remark or something serious. But if something serious, my best interpretation is that one can get ridiculous results from behavioural genetics using crappy methods. This is not news: One can reach arbitrary conclusions using crappy methods in whatever field.
  • A stronger implied argument is that most findings in the field are indeed crappy and that things like "genes that affect IQ/conscientiousness/linguistic ability?" are as bunk as the cantonese gene.
  • But even on this view, there are problems with the example. First, on a priori grounds, before looking at the concrete evidence, every complex behavioural trait that I am aware of is not mendelian (short of say, those that derive of sex, or certain illnesses perhaps). There is no gene for intelligence, or a gene that enables you to do complexity science. Second, we know there are non-native Cantonese speakers, and they most likely would not share the putative gene. If anything we would expect that if there is a spurious correlation between a trait and a gene, there are shared cultural factors behind (As discussed below in the chopsticks and tea examples); this would very unlikely be the case with a non-native speaker. So this example is quite bad to begin with.
  • What if we change the example slightly to make it better and say that there is a gene that makes you better at Cantonese (by a tiny amount), and that this gene is more prevalent among the native Cantonese population? Then it doesn't seem ridiculous! In fact, it could be true, a decade ago such genes were allegedly found (for tonal languages, not Cantonese in particular). In general, there is a research field trying to see if small, cross-population genetic differences that make some population more skilled at certain kinds of vocalizations can bias the evolution of the language (Which in turn leads to some interesting coevolution patterns). The point here being not that this is true, but that it can't be dismissed out of hand as silly.
  • In more practical terms, in twin studies, there are both monozygotic and dizygotic twins. For any given trait, for twins raised in the same environment, let's assume that both twins will speak the same language. Naively, if we have a sample of Western and Cantonese twins, it would seem like the language each pair of twins is perfectly correlated. Lo' and behold, it's genes! Not so quickly: If we assume that the correlation between both DZ and MZ twins is perfect (There is no pair of twins that speaks a different language) and we plug the numbers in the formulas to estimate heritability in the classical twin method we get that the effect of genes is zero! Of course! If the correlations do not change depending on the genetic similarity of the twins, it's unlikely that the trait under consideration has a genetic basis.
  • TLDR both in principle, and in practice, this is a confused thought experiment.
  • A better example or what the Cantonese tweet thread was trying to get at -causal identification is hard- is here, on "tea drinking genes", or this piece on "chopstick genes"