pipette
ESEspañol

Wednesday, September 23

1,915 new papers appeared. We read every one. These 20 are worth your time.

Each dot is one paper. The pink ones made today's edition. Point at any dot to see its title.

Today's edition

Picked for broad interest, a real step forward and honest claims, with no field allowed to take over.

  1. In-orbit operation of a programmable quantum photonic processor

    Here, we report a programmable quantum photonic platform operating on a nanosatellite, processing two photons in a six-mode universal integrated circuit.
    AI summary

    The authors built a programmable quantum photonic processor that operates on a nanosatellite in orbit. The device processes two photons through a six-mode universal integrated circuit. By programming different unitaries and tuning the photons to be indistinguishable, they observed two-photon interference. This demonstrates onboard generation, manipulation and detection of non-classical light. The work extends space-based technologies toward quantum-assisted computing in orbit.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big stepReal-world use
  2. Recursive self-improvement of AI research agents

    Together, these results show that an AI research agent can improve its own research efficiency through recursive self-improvement, and that these gains transfer to tasks and domains the loop never encountered.
    AI summary

    The authors built AIDE^2, a system where an AI research agent rewrites its own code and keeps changes that perform best. In an 8-day autonomous run, the system found seven improvements, including a new search policy and memory mechanisms. These gains worked on four held-out benchmarks covering machine learning engineering, heuristic algorithm engineering, and weather forecasting. On all four, the best improved agent matched or beat a strong human-engineered production agent. The authors show that an AI research agent can improve its own research efficiency through recursive self-improvement, with gains transferring to new tasks.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big step
  3. Rising water is an underappreciated driver of forest mortality

    Our findings identify rising waters as a widespread, underappreciated driver of forest die-off that can extend far inland, underscoring the need to reassess the impacts of climate change on the functioning, demography, and carbon-methane budgets amongst the world's most carbon-rich forests as seas rise and hydroclimate shifts globally.
    AI summary

    Forests can die from drowning, not just drought and fire. Researchers tracked over 260 million dead trees near U.S. coasts from 2012 to 2023 using aerial images and deep learning. More than half of tree deaths near coasts occurred in just 10% of forests, mainly in wetlands and low-lying areas prone to flooding. Death rates doubled around expanding lakes and inland wetlands, and tripled in low coastal forests with saltwater intrusion. Rising water is an underappreciated cause of forest death that may reach far inland.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintReal-world use
  4. Generative Neuromorphic Programming of Mammalian Cells

    Here we establish generative programming of mammalian cells through a neuromorphic framework that compiles desired behavior into DNA.
    AI summary

    Cells naturally compute analog responses through interconnected regulatory networks. The authors created a neuromorphic framework that compiles desired cell behavior into DNA. They built regulatory devices from RNA-targeting endoribonucleases to create multi-layer analog circuits. Their biomorphic neural networks predict circuit responses by learning from whole circuits containing each basic process. Their biocompiler software designed three new architectures that each worked in mammalian cells without manual tuning.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big step
  5. Valuing five centuries of unpaid breastfeeding in England

    We impute upper and lower bounds for the value of unpaid breastfeeding over five centuries in England to better understand its economic role over the course of economic development.
    AI summary

    The authors calculated upper and lower bounds for the value of unpaid breastfeeding in England over five centuries. They also estimated changes in breastfeeding duration using historical and bioarchaeological data. Unpaid breastfeeding was at least 0.1% of GDP and possibly peaked at 6.58% in the sixteenth century. This value exceeded spending on school and apprenticeship, making it possibly the largest human capital investment in early modern England. The relative value of unpaid breastfeeding gradually declined as the wider economy grew.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Peer-reviewed journalBroad interest
  6. Exo-LAMP: A Rapid Extraction-Free Isothermal Assay for Exosomal mRNA Detection

    Here, we report Exo-LAMP, an extraction-free isothermal assay for detecting placental exosomal KISS1 mRNA, representing the first isothermal approach for exosomal mRNA detection.
    AI summary

    The authors developed Exo-LAMP, a test that detects KISS1 mRNA from placental exosomes without extraction steps. The test captures specific exosomes with magnetic beads, releases RNA by heating, and amplifies it in the same container. The entire workflow finishes in under one hour and could be adapted for other RNA targets in liquid biopsies.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big stepReal-world use
  7. Autonomous Quantum Transport Measurements of 2D Semiconductors by an AI Agent

    Herein, we introduce an AI-driven workflow in which an AI agent performs multi-step, multi-day quantum transport measurements end-to-end.
    AI summary

    The authors built an AI agent that runs multi-day quantum transport experiments from start to finish. The agent plans measurements, operates cryogenic equipment safely, analyzes data, and writes a final report. They tested this on monolayer and bilayer MoS2 devices for up to six days. The agent found the conduction-band spin-orbit coupling energy in monolayer MoS2 and mapped a phase diagram in bilayer MoS2. The system, called FermiLink, prioritizes instrument safety and reliable results.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Preprint
  8. Bistability of compartmentalized autocatalytic systems at the dawn of hereditary variation

    Here we show that two mutually interacting autocatalytic networks can form alternative stable chemical states, maintained by weak cross-catalysis and switched by environmental change when enclosed in reproducing compartments.
    AI summary

    They found that two interacting autocatalytic networks can form alternative stable states inside dividing compartments. Rare compartments can switch to a previously unfavored state by chance and then spread when conditions change. This suggests early inheritance could have been based on chemical attractors rather than genetic sequences. The work points to a specific experimental signature: compartmentalized networks that remember past states but can still switch.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Peer-reviewed journal
  9. Multimodal Visual Sensing of Temperature and Pressure: From Spectroscopic Readout to Multiple Linear Regression-Enhanced RGB Analysis

    This work presents the first application of MLR-assisted RGB analysis for quantitative luminescence pressure sensing.
    AI summary

    The researchers created a material that can sense either temperature or pressure depending on how it is made. The material contains two different light-emitting centers that respond differently to heat and pressure. Using a statistical method called multiple linear regression, they turned color changes into precise temperature and pressure maps. This method improved temperature sensitivity from about 1% per kelvin to 8.8% per kelvin, and pressure sensitivity from 169% per gigapascal to nearly 721% per gigapascal. This is the first time this statistical approach has been used for pressure sensing with light-based imaging.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big step
  10. Dense Packing of Tetrahedra in Cylinders

    Combining numerical simulations with exact analytical theory, we classify the densest configurations into two continuous, mathematically tractable families: chiral Tetra Helices and achiral Dimer Chains.
    AI summary

    The authors studied how pyramid-shaped tetrahedra pack tightly inside round tubes. They used computer simulations and math to find two families of densest arrangements: spiral Tetra Helices and straight Dimer Chains. They show that the famous Boerdijk-Coxeter Helix naturally arises from this tube confinement.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big step
  11. Multidirectional Colossal Polarization in a Centrosymmetric Crystal via Long-Range Ion Displacement

    Herein, we demonstrate that colossal polarization occurs by long-range ion displacement even in the nonpolar proton conductor Na2MnN(CN)4·3.25H2O (1·H2O).
    Peer-reviewed journalClaims a big step
  12. Dr-LiSA: Direct Radar-Lidar Scan Alignment for Localization

    This paper introduces Dr-LiSA, a first-of-its-kind direct method for localizing 2D spinning radar intensity measurements in against 3D lidar maps.
    AI summary

    Dr-LiSA is a new method that matches 2D radar scans to 3D lidar maps to find position and orientation in 3D space. Radar works well in bad weather, while lidar makes detailed 3D maps in good conditions. Past methods could only find position on a flat plane and were less accurate than lidar-lidar or radar-radar systems. Dr-LiSA uses a learned model to predict what radar would see from a lidar map, then matches predicted and real radar scans directly. It beats earlier radar-lidar methods in flat-plane tests and matches top radar-radar accuracy across over 90 km of road data.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big stepReal-world use
  13. Simulation and Analysis of Solute Transport in Multi-Lymphangion Lymphatic Vessels

    Overall, the proposed PBS framework provides a first quantitative basis for solute transport modeling in the LS and several concrete application scenarios for MC in this underexplored domain.
    AI summary

    The lymphatic system, which controls fluid balance and immune monitoring, has not been studied for molecular communication uses until now. The authors built a particle-based computer simulation to model how substances move through connected lymphatic vessel segments separated by valves. Their simulations show that valves cause solute concentrations to come in bursts over time, matching what is seen in living tissue. Within one pumping cycle, valve timing controls when peaks occur, while diffusion and receiver location control how sharp those peaks are. This framework offers a first quantitative basis for modeling solute transport in the lymphatic system and possible applications for molecular communication.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Preprint
  14. Learned Human Aortic Morphodynamics: A Dynamical-Systems Model of Post-EVAR Sac Remodeling

    Recovering the equations that govern remodeling, rather than tracking its instantaneous state, therefore provides a foundation for clinical models built from the sparse records that routine imaging surveillance actually produces.
    AI summary

    The authors studied how the human abdominal aorta changes shape after a stent graft is inserted to repair an aneurysm. They analyzed 220 CT scans from 100 patients at two international centers, reducing each scan to two measurements: surface area and shape curvature. Using a sparse identification method, they found mathematical equations describing how successfully healing sacs and stable sacs change over time. The learned equations allow earlier prediction of patient outcomes than looking at single snapshots alone, even when follow-up data is incomplete.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintReal-world use
  15. Generating protein hydrogels with customizable stress relaxation behavior via deep learning-driven entanglement design

    Here, we introduce TangleDiff, a deep learning framework for the de novo design of homodimeric entangled proteins with programmable features.
    AI summary

    Protein hydrogels are useful materials for growing cells in 3D because they can relax stress under strain. The authors created TangleDiff, a deep learning tool that designs new entangled proteins with specific features. TangleDiff succeeds over 70% of the time in designing foldable entangled sequences, far better than existing models at about 1%. The team used TangleDiff to design nine protein pairs with specific binding energies; seven formed hydrogels with stress relaxation matching the design targets. This work provides a general method for designing entangled proteins, enabling new types of biomaterials.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Peer-reviewed journalReal-world use
  16. Beyond Natural Language: An Agent-Native Language for Autonomous Science

    We introduce Lara, a machine-checkable language and protocol for checking and revising support for research claims.
    AI summary

    The authors created Lara, a machine-checkable language for verifying research claims. Lara turns research arguments into executable artifacts that enable automated validation. Authors explicitly declare claims, evidence, assumptions, and limitations in a Lara program. A deterministic checker assigns each claim one of four statuses, including 'gap' for incomplete support. Case studies include empirical review, philosophical debate, and loss of support when an axiom is withdrawn.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big step
  17. Rachel: A general-purpose language model directs and revises retrosynthetic routes

    Within Rachel, a general-purpose LLM coordinated successive chemical choices and revised its strategy as earlier decisions reshaped the remaining problems.
    AI summary

    The authors built Rachel, a system that lets a general-purpose language model direct chemistry without fixed search rules. GPT-5.5 found complete routes for 111 of 120 standard targets and 24 of 25 difficult targets, with no reference solutions given. The difficult targets came mostly from studies published after GPT-5.5's knowledge cutoff. When the authors replaced the language model's decisions with fixed rules, success dropped sharply to 6-15 of 120. The model revised its strategy as earlier choices changed the remaining chemical problems.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Preprint
  18. Data center cooling choices shift water impacts across the grid: An integrated water-energy model for sustainable data center development

    Air-cooled data centers halve total water consumption relative to evaporative cooling, but increase electricity demand and raise indirect water consumption by one-third, shifting the water footprint from the facility to generators.
    AI summary

    Data centers use water directly for cooling and indirectly through electricity generation. The authors built a model that tracks hourly electricity use and monthly water depletion by specific river basins. They tested different cooling setups for data centers in Michigan. Air-cooled centers cut total water use in half compared to evaporative cooling, but raised indirect water use by one-third. This shifted water impacts away from the facility to power plants in other regions.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintReal-world use
  19. Relegation, promotion and the components of scoring in water polo

    Fitted to a new dataset of 940 matches, the model identifies the directly relegated club from the ninth round, before the league table does, predicts the play-out field better than the table, and shows the play-out itself to be close to a coin toss.
    AI summary

    In many European water polo leagues, avoiding relegation matters more than winning the title. The authors built the first Bayesian hierarchical model of water polo scoring. The model splits scoring into even-strength, man-up and penalty parts. Using 940 matches, it spotted the relegated club by round nine, before the league table could. Promoted clubs start weaker than others, except at drawing exclusions.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    Preprint
  20. First Observation of a Polar Coronal Hole-like Fast Solar Wind Stream in the Sub-Alfv\'enic Solar Corona: an Analysis of Turbulence Properties

    At heliocentric distances of approximately 10 solar radii, with speed mostly above 400 km/s, this interval is a first example of "polar coronal hole-like (PCH-l) fast" solar wind observed in the sub-Alfv\'enic solar corona.
    AI summary

    This is the first time such fast wind has been observed this close to the Sun, at about 10 solar radii. The turbulence in this wind was already well developed, strongly sideways, and highly imbalanced. These findings give new limits on how solar wind turbulence grows in the lower corona.

    Written by Kimi K2.5 from the abstract only. Jev checked every sentence against the authors' words. See where each sentence comes from

    PreprintClaims a big step