Cognitive Revolution has an interesting interview with Gemini Robotics on recent developments in robotics. Unclear to what extent the fact that frontier foundation models are working for physical tasks disproves my theory that shape-rotator architectures are necessary for correctness and creativity. Maybe instead of replacing embeddings, concept shapes will be produced by them and then stored in external memory, like a specialist gradually developing their mental model. Then the way you would check the validity of some idea is by loading the relevant concepts and then outputting in which ways they align and their relative goodness of fits (initially, this could be two concepts and some particular input dimension, and then the output would be the goodness of fit there).
Jinay releases Soarxiv, an embeddings visualizer for arXiv papers (via Divya Venn).
Yondon Fu on the future of customized gaming. Although I feel like if the technology that enables this is possible, then AR gaming will also be possible and way better. ARX-Han has an interview where he shoots the shit with Pedro Gonzalez (charitably speaking, that’s talking about culture) which is entertaining and includes a mention on the potential of AR; I include this as evidence that this AR future is so obvious that most people who are terminally online are predicting it. Anyway, it’s interesting to note that Yondon’s example of a customized AI generated character is actually using the model of a real VTuber. The link to reality isn’t just vestigial, it’s integral.
Mutual Understanding podcast on the TPOT scarf debate. Very interesting how an abstraction of this idea, the morality of violating several norms in service of personal gain if no one is harmed, is tied to so many other issues. Not just agency, but also the morality of piracy, the nature of romance (speaking of which, I recently read All Things Are Too Small, which was a good reminder that Substack is primarily a destination and not origin for discourse), high versus low-trust societies and clannishness, etc. For my personal take on the morality of this situation, I think the circumstance was fine, posting it was unadvisable, and condemning it is correct. This is based off the fact that consequentialism is the purpose of morality, that one should act according to virtue ethics when second-order effects are non-negligible, and that status should be determined by the categorical imperative. Possibly related, Robin Hanson has a post on the complexity and power of norms.
The latest issue of the LRB has an interesting issue on the recent happenings in the Catholic Church, and it’s relation to American politics, including the interesting fact that apparently Steve Bannon outperformed the prediction markets.
Anticipating the Unexpected continues their coverage of the India-Pakistan conflict, including on the performance of the J-10C, the retaliatory stance being adopted by the Indian government, and different online narratives that came from within India. Samo Burja also has a podcast on this topic, but I’m not sure how seriously to take it: for the most part it’s not predictive, just post-hoc analysis. Also, whichever AI editor he’s using doesn’t seem to notice that Samo will variously say India or Pakistan when he means the other.
For more on geopolitics, Joe Walker and Martin Sustrik both have linkthreads.
Republic of Letters with numbers on the extent of how women have captured the publishing industry. To be honest, a lot of these numbers are actually pretty low: 58% of literary agents and 78% of publishing house employees. I was actually expecting higher than 80%.
José Luis Ricón on retinoids, probably a response to this LessWrong post. The picture I’m getting is that retinoid use is more situational, something you could try if you have a lot of acne scars or acne. Beyond that, it seems it’s more up to personal risk-tolerance for unknowns.
Stephen Malina linkthread, on different rationalist-adjacent topics.
