2026-05-15
king of whitewater
Approved Podcast on the Keytruda’s very crazy development story (some additional commentary).
Ben Southwood with a self-profile of his intellectual development from dogmatic libertarianism to a progress-oriented liberalism. He includes a very interesting quote, that “markets and prices, and the incentives and information they give us, can’t really be what’s important about capitalism. Instead, the importance of capitalism is that successful central planners (like Jeff Bezos and Elon Musk) get more and more capital to plan”. This is something I’ve been considering recently, having finally caught up with the current publication progress of Illumine Lingao, which I now understand as the Industrial Party’s trading agenda for legitimacy with the CCP, by providing an answer to the Hayekian critique of central planning: that the knowledge problem is actually irrelevant, because if central planning can direct resources towards the most advanced productive forces of society, then long-term wellbeing will naturally increase as a result of the diffusion of technological progress. As a techno-optimist, this is actually quite a hard position to directly refute. The best I can do is an indirect critique, that there remains the issue of ensuring the planning system can maintain its own quality and dedication to its core mission: China’s current struggles show that limits exist to export discipline as a means of obtaining non-Goodharted feedback, necessitating recursive anti-corruption campaigns which continually inject political instability into the system. Within the techno-optimist premise, if there is a justification for liberalism, it is that no single party should have so much power that they no longer feel the need to compete to justify themselves. Of course, meritocratic liberalism also has its own potential failure cases and means of subversion, such as the stagnation that arises from the desire for existence without effort and an end to competition, which leads to rent-seeking, nepotism, and other forms of low-trust clannishness. But generally, it seems to me that, while central planning might be short-term and resource-efficient, competitive planning remains the superior option for overall outcomes in the long-term. But then that is itself something that cannot be set by fiat, and will necessarily be determined by natural selection.
Anthropic comments on geopolitical AI competition.
Scott Alexander defends continuing trendlines as a means for predicting AI progress, which is entirely reasonable, but I’m rather confused with the justification that “the best way to predict this is to fully understand the process generating the trend [but] what if you don’t fully understand the process?” It seems perfectly functional to use curve fitting on exponentials as a heuristic for estimating lower and lower bounds, with both flattening out and continuing forward as variations of “nothing ever happens”, and Lindy’s Law as a workable middle ground. But this should not be confused with actual understanding, for which difficulty is hardly a valid excuse for dismissal1. For example, one way to improve the lower bound is to ask producers themselves how things are going: presumably if one asked solar panel manufacturers about costs, that would allow one to predict solar deployments more accurately than the IAE2; likewise, if one asks Chinese women3 how many children they want to have, a different picture emerges than the one that is presented by the United Nations. For predicting upper bounds, there are physical limitations, and in the middle, one can predict the rate of progress through understanding of current bottlenecks and hard problems, weighing their expected difficulty versus the amount of talent and resources which are being thrown at them4. These are all factors for creating a robust portfolio of potential outcomes, based on a mechanistic model of the world which can be improved in real-time as additional information is revealed; in contrast to curve-fitting, which can update but never learns.
Amritanshu Prasad against data scrubbing as a means to prevent AI from engaging in undesirable behavior.
Benjamin Lyons on intelligence as the ability of the system to coordinate itself to solve problems, not only through reasoning, but also instinctive response.
Vincent Huang overview of global supply chains for major commodity products.
Maja has an article on hustling for a job, which I view as a sign that people are finally starting to wake up to the fact that hiring has broken, in that jobs are no longer obtaining through formal hiring channels. In any case, telling people to learn to sell themselves seems like a rather bad development, since there seems like there are inherent inefficiencies to hiring people who are very good at hustling for jobs5, essentially handing your recruiting budget and process to the hustlers themselves.
Abundance and Growth Blog linkthread.
Alexander Kruel linkthread, which includes an interesting survey of a Physics Magazine readers on their positions on various controversial topics in physics.
It seems to me that if one spends a lot of time trying to justify why people should never work for AI labs, it might be easy to forget that there are significant benefits to developing a mechanical understanding of the underlying processes which are driving forward progress, one of which is being able to construct a model which leads to more accurate predictions.
Although, I actually think solar prices are very likely to stop falling soon, due to a combination of material constraints and anti-involution policies, though fortunately panel prices are not the only factor affecting their rate of deployment.
According to my current understanding of why Chinese fertility in particular is so long, it’s that while in America there is a division between liberal and trad culture in fertility outcomes, in China traditional female identity is also heavily anti-natalist. By which I mean, according to traditional conceptions of lineage, descent is entirely carried by the male line, which means that a woman’s children only ever temporarily belongs to her. Hence the difference between male and female fertility desire for children, and rising bride prices despite government attempts to cap them.
Of course, unknown unknowns do exist, and should be accounted for, but actually the more resources a problem is currently getting, and assuming that participation is open to outside participation, that itself serves as evidence for how hard different problems actually are and that there may be fewer unknown unknowns than might typically be expected.
Conflict of interests aside, it is a well-known result that if one selects on two independent variables, the typical result is a negative correlation for these two variables within the selected for population. Therefore, this might be a good strategy for hiring marketing or sales staff, but perhaps is not the best idea for general purpose hiring or acquiring technical talent if the goal is efficient conversion of capital to capabilities.
Just want to say again that I appreciate you!