John PSmith has another excellent book review, this one analyzing The Hard Thing About Hard Things by Ben Horowitz of a16z and noticing the parallels between honor culture and the startup mentality (Edit: similar comparison made in this episode of Takeover Pod).
Scott Sumner on AI (lack of) creativity1, where he has a sort of Platonist take that’s there’s some realm of truth which is currently inaccessible to LLMs. This seems kind of misguided to me. There’s a section where, on the evil-good feature, he says that “this is another piece of evidence that scientific, ethical and aesthetic knowledge are real, and are part of a unified whole.” Personally, I see it more as evidence that within human verbal culture, those things are regarded as part of a unified whole, but it’s unclear if this is actually the case (or since, it’s hard to say what that means for something intangible, we can limit discussions to whether this is universal). This is actually an argument to continue AI capabilities development, because we are likely to encounter other intelligences eventually either way, and the likelihood of them being aligned with us is much less than the likelihood of alignment with LLMs trained on human culture.
Some stuff on the fusion of mind and machine by Anu Atluru and Tina He. I find these fun to read, but personally, I think it’s more or less impossible to accurately predict how things will play out at a system level from within the system itself. Other interesting AI stuff today: Anson Ho on the METR task-length benchmarks, and the limits of benchmarks in general; Asterisk Mag discussion between Ajeya Cotra and Arvind Narayanan; Zvi with more on 4o sycophancy; Lars Doucet continues his short story serial on trading with ants.
New Emergent Ventures cohort from India. It’s interesting to see how the choice of what link to represent someone changes according to the cohort location. Low ratio of blogs (or even Twitter) from India, unfortunately.
Theo Jaffee on writing well-reasoned and intentioned takes on controversial issues: with trans-issues as a case study (via Tracing Woodgrains, at the same time the poly question is also flaring up in TPOT again). On that topic, there’s also Yascha Mounk interviewing Claire Lehmann.
Matt Yglesias wonders why large companies aren’t publically criticizing bad Trump policies. It seems to me that the obvious reason is that these companies want to avoid the unenviable situation of both parties hating them. Assassination of Big Pharma executives get celebrated, and antitrust has spent the last decade trying to break apart Amazon and other tech companies. If Trump decides to retaliate after any criticism, will Democrats protect them, or will the Gray Lady speak up on their behalf? Seems equally likely to me they would celebrate their demise.
Father Karine has a cool piece about growing up in rural Appalachia, which to me is evocative of the PSmith review of Sick Societies as a discussion of tradition and Chesterton’s Fence; it’s usually there for a reason, but that reason might be overfocus on some particular metric that leads to a local optimum.
Asimov Press on the history of the centrifuge and it’s scientific fruits. It’s interesting, because a quick search indicates that centrifuges are not manufactured exclusively in China, but are still manufactured in places like Europe, Japan, and America. I expected that multipurpose devices would benefit especially from agglomeration effects, so I suppose this means that centrifuges are highly specialized for their different use-cases.
Rebecca Lowe philosophy linkthread.
Nicholas Decker economics paper linkthread.
Left a comment on this, about how I think AI creativity probably requires shape-rotator architectures.