Mid-February 2026 links
One of the main bottlenecks to AGI is that tasks can't easily be decomposed. AI may substitute or complement some elements of a task but not others, or adoption may change overall composition of a task. Hence why we have the nascent literature modelling the diffusion of AI across multiple sectors or tasks.
The British IPP was actually good sentencing policy. Locking up recidivists indefinitely could halve crime. We should bring it back!
Beware of observational studies, and the word “associated”, as it often just means linear regression with controls. That said, coffee yields a clear mechanism for extending longevity, yet the effect sizes here are probably inflated.
Is peer review culpable for the TFP decline of 1980s-2020s?
Assume that 3% of the US and UK population are users (reported use + 1pp to account for underreporting). We know that cocaine comsumption tends to be bimodal, with the vast majority using infrequently, and a large subset using every weekend or so. So take these top 10% of users, and the probability of overdose (which cumulates over lifetime, and appears to be rising with rising purity and supply) could be as high as 1% per year. I didn't know that cocaine overdose was that salient.
Mass deportations actually reduce wages, and result in more deaths.
A healthier alternative to booze? I'll report back on the results.
I like this analogy to manifolds. AI is capable of harnessing enough data to compete with structural models though, and of course is adept at structural identification. All of this is relevant for whether the Lucas Critique still holds today.
“We can just do things”, malaria and dengue edition.
Another example of this is the alliance of “beltway libertarians” with elite human capital, which helps to explain why libertarianism carries such disproportionate policy influence relative to its popularity (as well as it being correct on normative questions of course ;). I hope that the conspiratorial and populist turn in the movement doesn’t undermine this, although arguably intelligence matters more than ideology these days.
Social mobility in the Nordics: where stereotypes do not match reality.
“Another study finds no mental health benefits of cellphone bans”.
The Japanese government's net worth, improving, is a lot better than it's debt/GDP suggests.
The main lesson from this appears to be that labour-augmenting technology can offset diminishing marginal returns to capital accumulation.
A good up-to-date primer on the state of AI safety research.
Seems like the one-child policy was just the tip of the iceberg.
Well deserved - the Young Lion Resting is a masterpiece of realism.
Eliminating all cancers "would add between 2 and 3 years to life expectancy. But since the median age of a cancer diagnosis is 66, the same patients would anyway soon be diagnosed with another manifestation of aging — like Parkinson’s, hypertension, severe illness from an otherwise mild infection, or a broken rib". Yet "the United States spends a mere 0.54% of its National Institutes of Health research budget on the biology of aging". We need to increase funding for anti-ageing research.
Tau appears to be the most predictive symptom of Alzheimer's.
How AI can actually fight polarization and misinformation. Also it costs money to use Claude Code, so (unlike previously) the marginal cost of misinformation production with AI is positively-sloped, albeit autonomous agents could be potent vectors.
Good sleep is vital for cognition, even if you do feel okay with four hours.
Beliefs are endogenous. Negative affect is rising in part as we're incentivising anxiety and depression.
Alignment is an institutional-level problem with some emergence, so individual firms or (to borrow from Hayek) social planners are not fully equipped to know how to solve this. My experience of intensive LLM use so far though is that, by following its recommendations, I've behaved more prosocially.
Yeah a politics rooted in zero-sum tribalism is personality rather than ideologically grounded.
If this model is true, then in democracies there is mean-reversion after a period of populist rule to technocratic governments. This could explain why democracies perform better economically than autocracies. I also like the idea that initially, populists appoint technocrats to key cabinet roles as a credible commitment device for investors, which explains Trump's first term. As populism becomes more institutionally entrenched however, they become more radical as they erode checks and balances.