Construction Physics, on how robot dexterity, is still an unsolved problem. Related, Chris Best had a discussion with Mills Baker on LLMs and consciousness, where they mention an interesting idea that LLMs represent the extent to which wordcels, given enough raw computation and memory, can mimic abilities nominally belonging to shape-rotators. The idea has some interesting implications, such as whether the current lack of discovery by LLMs has anything to do with how good ideas feel like (to me) concepts physically fitting together, a sort of filtering method that reduces the complexity of idea generation immediately to idea verification. Or to what extent LLMs can be adapted to robotics, or whether low-latency robotics might require a neural network architecture where shape-rotating is the first-class citizen, more natively suited to modeling physical objects and space (edit: for example, if transformers can be thought of as Graph Neural Networks with tokens as nodes, then this would be some equivalent to the transformer but where edges are the first-class citizens. If I’m not misunderstanding, this is basically how AlphaFold works, and that feels very shape-rotating to me).
Luke Drago and Rudolf Lane publish the Intelligence Curse, on how to prepare for fundamental disruptions in human society and economy following non-foom but nevertheless transformative AI. In my eyes, the key tension is between the need to guard against enhanced-capability negative externalities (eg. bioterrorism), and having sufficient diffusion of technology such that people can credibly threaten to exit. On our current path where we solve everything with centralized controls, the latter will remain hopelessly politicized, leaving people with no answer to marginalization or oppression.
Statecraft interview with David Lebryk at the Bureau of the Fiscal Service in the US Treasury, and how the optimal amount of fraud is non-zero. Earlier episodes felt rather parochial, but recently these have been really good, making you feel like you are learning generally applicable principles reusable across different domains and organizations.
Crystal Duan writes about being single, following a tweet by Noah Smith on how Gen Z is dropping the ball. Contrary to the belief that increased singledom is essentially just jobless gamers living in their parents’ basement, this is probably a fairly common failure mode. First there is an absence of formative romantic experiences during high school (due to having controlling parents or maturing relatively late sexually) and university (by entering a gender-bifurcated field, like engineering for guys or apparently writing for girls). This is paired with some combination of anxiety, autism, or bad experiences, which results in one being closed off and oblivious to interest cues. After figuring out other ways to deal with desire for social interaction (work, hobbies) and sex (porn, smut), one ends up forgetting that dating is an available option. After a prolonged period, there is strong psychological resistance to restarting: since you have no idea how you are supposed to behave within a long-term relationship, even if you get one. This makes you feel that you are likely to fail, so only the best-suited matches are worth any effort, but this results in a scarcity mindset that further raises pressure. Without strong self-introspection and extremely high agency, this is probably inescapable, so my personal opinion is that they should teach and encourage practice relationships in high school to prevent this particular failure mode.
Henrik Karlsson on failure to realize your deep compatibility with people that you already know.
Thomas Pueyo on the purposes of fashion.
