What I have been reading: mid-September update
After a lengthy hiatus, I am once again active on X. My primary motive for using the platform is simply to find and promote content which I find interesting. Despite the perceived debasement of the platform under Musk, X is still the focal point for sharing the latest papers and articles, and for topical debates in real-time amongst some highly intelligent individuals. However, there are limitations to using my feed as a personal library; it gets crowded over time, so it becomes impossible to remember everything. Additionally, the algorithm now deprioritises links, so the marginal cost of disemminating articles has risen. Hence, to take inspiration from Marginal Revolution and Scott Alexander, I will now post bi-monthly updates on here, summarising what I have read.
Smartphone bans in the classroom generate a modest improvement in academic performance. However, the average increase is only 0.086 standard deviations. We must interpret this finding alongside the potential opportunity costs of this intervention (collecting phones consumes time and human resources after all, which could be diverted to more effective interventions).
If declining an invitation, always say so explicitly. Do not insult the host's intelligence by equivocating with "maybe". I admit, I am biased. This paper fits my priors well! This is no different to when you get “I’m busy on x day” after you ask someone out on a date.
Pensioners are doing very well from the welfare states of France and Britain, whilst living standards are roughly stagnant for the rest of us.
Do not believe the recent claims that AI progress has stalled. LLM’s can now reason through some of the toughest maths problems in existence. What are the odds that the proof of Fermat’s last theorem, or the solution to the P=NP debate, will come from AI? Arpit Gupta is right on the challenge to the Lucas Critique. Is this the start of a Kuhnian paradigm shift in macro?
The Brahmin left drives support for the Labour Party in Britain.
More on AI progress. The bottleneck is the physical world, not the capabilities of LLMs.
Observations of TRAPPIST-1e
Young liberal females saw the greatest rise in depression throughout the last decade relative to other groups. Correlation is not causation, yet this does show that feminism does not guarantee happiness for women.
One of the beauties of capitalism is that there are multiple status hierarchies one can compete in. This is underrated as a source of social stability.
“Nearly half of US federal judges attended a crash course in economics”. They “sided against regulators more often, and imposed harsher sentences”.
Liberté, égalité, radioactivité. In the 1980s, France built nuclear reactors at a pace faster than China is now.
GPT-5 Pro is able to perform novel maths, when guided by a maths professor.
It is incredibly rare for a person to have high IQ, conscientiousness, and emotional stability Talent is multidimensional, yet it is exceptional to exceed across all dimensions simultaneously. This is why I am pro-natalist. Let's make a key resource in the advancement of human civilization a little less scarce.
AI alignment is a laudable goal, but what if the humans using the AI have malicious intentions?
The effect of social media deactivation on users' emotional state before the 2020 US election. A great example of results being statistically significant but not large; merely a change of 0.03 to 0.05. In other words, from the 50th to the 52nd percentile. The contribution of social media to the mental health crisis is a rounding error at best.
A thread on AI and job losses. Whether AI is a complement or substitute to labour as a factor of production depends on the employee and the task at hand.
“We find that going from the most left-wing authored estimate of the taxable top income elasticity to the most right-wing authored estimate decreases the optimal tax rate from 77% to 60%." In this case, “optimal” is the revenue maximizing rate, assuming (based on the Mirrlees model) zero weight in the social welfare function on a marginal dollar received by a top earner, rather than Pareto efficiency. This relative uniformity amongst economists with heterogenous political preferences is testament to the quality of the profession. Pervasive social and cognitive biases matter less to the conclusions of research than one thinks. Perhaps mainstream economists are right more than we give credit to them for?
Is there a link between political extremism and neuroticism? Timely given the assassination of Charles Kirk. My priors are that isolated, sporadic acts of political violence follow a Poisson distribution. It is well known that the timing of wars follows this pattern. If neuroticism is uniformly distributed across the political spectrum (a reasonable assumption), then “the left” is less to blame than human nature, and a good dose of randomness.
Is P-hacking less a function of motivated reasoning, and more a function of benign ubiquitous human error which aggregates to distort the findings? Incentives are key here. Journals are well-known for incentivising the publishing of statistically significant over null results. Change the incentives, and do you change a key parameter (in this case, attention) in the production function for research? Or perhaps we may see a composition effect (more null results relative to significant ones) which eliminates this observed asymmetry?
King’s College London has scrapped the DEI litmus test for hiring job applicants. Left-wing wokeness is in retreat, yet with the Trump administration demanding affirmative action for viewpoint diversity within academia, this is barely a bittersweet victory. In particular, there will be incentives to game such a system. Research will no longer become a quest for pursuit of knowledge and the truth, and more a quest of finding oneself on the right side of the quota. Freedom of thought and speech is impossible in such an environment. We are making the mistake of substituting wokeness for an uglier phenomenon, just as we achieved a tentative victory in the fight for academic freedom.
Even superforecasters underestimated the speed and scale of AI development.
A thread on the sex recession. Is it a coincidence that this decline began just as the concept of sexual harassment was in its ascendancy? The risk of being subject to a costly civil lawsuit clearly imposes a transaction cost to finding a sexual partner.
Did alcohol facilitate the social and cultural evolution of mankind?
Populism as a conspiracy theory; a strategy of promoting a false alternative reality. “Alternative realities are endogenously conspiratorial to resist evidence better. Populists, to leverage or strengthen beliefs in the alternative reality, enact harmful policies”. In other words, low human capital is endogenous to populist movements across the political spectrum.
There is also progress in AI detection. Do not write through chatbot, write through your own words.
Regular reminder that robustness is still an issue. Do not update your priors solely on the basis of one paper. Do not accept the results of a single paper at face value. Note that I am not promoting radical skepticism either: you have no alternative but to engage your brain when digesting empirical research.
The Cambrian explosion may have started millions of years earlier than thought.
Derek Thompson on the state of the current health discourse. Despite the nascent revolution in biotechnology, do not neglect the foundational “bread and butter” of diet and exercise!
AI can be trained by another AI autonomously. Once again, model autonomy and AI progress is not the problem.