2025-09-29
yoshimi battles the pink robots, pt. 1
Liv Boeree interviews Emmet Shear on his ideas around AI alignment, and his worldview in general which is an aggregation of all the best and most useful ideas of TPOT (although it’s unclear to me which side owes more influence to the other). Given his mention that corporations behave like cancers, it’s worthwhile to listen to the interview in Live Longer World with Robert Gatenby on cancer from an evolutionary perspective, which also includes Michael Levin and his view that applying motivations to cells is useful for understanding their behavior. I’ve been thinking a lot recently about cancer analogies as applied to society, because usefulness in one direction implies that applying the metaphor in the other direction should also be useful1. If cancer forms when cells start behaving selfishly due to no longer viewing themselves as part of the collective, then many of our current societal problems can be explained as the breakdown of society as a collective organism into mutually hostile groups. There’s some relation here to the debate over to what extent inequality is a problem2, because as social mobility becomes harder, your incentive to participate in this system which does not serve you declines. This is an argument for meritocracy as a good way to keep things aligned, because it prevents the most capable from defecting, while the discontent of the incapable causes no harm. But as organisms age, their cancer rates increase, and quantity has a quality all on its on.
Razib Khan describes the western worldview as a fusion of ancient Greek and Jewish culture.
Kai Williams on the potential for AI to improve the prediction of earthquakes.
Cameron Wolfe with an overview of reinforcement learning.
Asimov Press on the history of cataract surgery.
Sam Matey on the fallacy of conflating naturalness with goodness. On that note, Cremieux on the potential for universal access to GLP-1RAs on ending obesity.
Because despite my utilitarian beliefs, which tell me I should be contributing to society and reproducing, my revealed preferences indicate that I don’t actually intend to work or marry; combined with wanting to live forever, at best I am a benign tumor. I wonder if the reason many rationalists and AI researchers reach first for control as a means to align AI (although I think it’s getting better) because the coherence worldview of alignment is in some sense implicitly judgmental of their lifestyles, which might be in service of malignant systems or transformative towards benign tumor-ness.
Maia Mindel has an interesting essay on this topic, although personally I don’t agree with many aspects of it. For example, I don’t think it’s clear that declining fertility should mean that income inequality between women is increased (by higher motherhood penalties for white-collar workers), because fertility is declining most among less educated women. Also, Matt Bruenig on antitrust, and Noah Smith reviewing Doughnut Economics. Anyway, I have a pretty good intuition pump for why our current level of social mobility is actually pretty good, which is the fact that agency and optimism lead to better life outcomes. The very fact that these worldviews are useful implies that it is “true”; if it were false, then such beliefs would have you constantly wasting your energy to no effect. Therefore, optimism is not only instrumental, but also justified.