Links #1

Intelligent spiders, simple models of spacetime, mathematics of x-risk, and more

How much do you believe your results? Highlights the importance of experimental error when choosing which interventions to scale. High-variance experiments can be virtually useless from an impact standpoint.

The Returns to Government R&D: Evidence from U.S. Appropriations Shocks by Fieldhouse and Mertens. Suggests that the U.S. is massively under-funding R&D.

Etched is making specialized chips for language models. My hope is that in the future every device will have several language models on it that are fast, private, and safe.

Sean Caroll recently gave an interesting seminar on his work attempting to derive spacetime from a simple quantum mechanical model that doesn’t assume the existence of space itself. I think this line of research is exciting, it’s possible that gravity (and everything else) is a consequence of a network of interacting quantum systems. See also the “It from Qubit” approach from the Simons foundation. EDIT: Entropic gravity seems to be a related approach explaining gravity as a result of quantum-level interactions.

In space news, Viridian Space has released a video explaining their approach to refueling in orbit and accessing VLEO. Andrew Cote has a nice thread explaining how railguns could be used to send payloads to space for $10/kg.

The Portia spider displays a shocking amount of intelligence despite having a tiny brain. Research into how it uses its limited computational resources are fascinating.

Viable offspring derived from single unfertilized mammalian oocytes. A step towards human parthenogenesis!

Acoustically-Targeted Measurement of Transgene Expression in the Brain. Uses ultrasound to release a protein from a specific part of the brain into the bloodstream, allowing them to confirm the insertion of the gene that codes for that protein.

Related: Openwater is using ultrasound interference patterns to better image the body and target ultrasound therapies.

Repair Biotechnologies is developing therapeutics to degrade atherosclerotic plaques. Heart disease is the #1 killer in developed nations and there should be a lot more companies focused on it.

Related: Verve Therapeutics is working on gene therapies to reduce cholesterol in the blood.

Induction of a torpor-like hypothermic and hypometabolic state in rodents by ultrasound. Seems useful for figuring out how to put people into a state of suspended animation. Could we do the opposite and make people more alert using ultrasound zaps?

Feasibility of mapping the human brain with expansion x-ray microscopy. This would be expensive, but doable. It seems that progress on mapping the human brain is moving much faster than actually simulating it, suggesting that simulation is the real bottleneck to digital minds. The rest of Logan’s blog is excellent as well, see for example his Global Highlights in Neurotechnology.

ZipIt! Merging Models from Different Tasks without Training. Demonstrates some new techniques for combining deep learning models trained on different tasks. This is valuable from a practical and safety standpoint; rather than train a single model to do many tasks, we may be able to train many specialized models and combine them. This is both easier to do and safer, since each component can be checked independently for safety.

Three mistakes in the moral mathematics of existential risk

The heritability of fertility makes world population stabilization unlikely in the foreseeable future. This makes me feel a little less worried about fertility collapse, but I remain skeptical. Regardless, we should still be testing and implementing pronatal policies in case fertility continues to fall.

The Algorithm for Precision Medicine. Offers a beautiful story and a vision for how precision medicine could actually work.

Macro Oceans is using kelp farming to produce chemical products. This is neat because kelp doesn’t require land or freshwater and could help capture carbon.

The Universe of Minds. This one is a little abstract, but thinking about different possible minds and their relationships seems useful. Some ethical theories may take into consideration the opinions of minds that don’t currently exist.

I just came across The Matrix Cookbook for the first time. It seems to have a lot of useful linear algebra that is hard to find elsewhere.