pipette
ESEspañol

Thursday, September 24

1,880 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. Large-scale esophageal cancer screening through noncontrast computed tomography and artificial intelligence

    Here we developed the Esophageal AI-Guided malignant Lesion Evaluation (EAGLE) model to detect precancerous lesions and cancer from chest NC CT, a task historically considered impossible.
    AI summary

    For cancer detection on existing CT scans, the model achieved 90.0% sensitivity and 98.5% specificity. In real-world low-dose screening, it reached 99.94% specificity, and calibration reduced false positives by 72.7%. The authors conclude that EAGLE has the potential to serve as a scalable tool for early esophageal cancer screening.

    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 journalClaims a big stepReal-world use
  2. Dissecting land subsidence along United States coasts through hazard characterization and exposure disparities

    Here, we present the first national-scale characterization of land subsidence hazard and exposure of population groups and infrastructure systems across 20 United States coastal states, explicitly accounting for spatial dependence.
    AI summary

    The study mapped land subsidence across 20 U.S. coastal states using over 190 million satellite and ground measurements. Nationally, 71.5% of coastal areas sink faster than 0.1 centimeters per year, with the Gulf Coast showing the highest rates. The researchers classified subsidence into low, moderate, and high hazard levels. Disadvantaged communities and critical infrastructure face disproportionate exposure to this hazard. The findings aim to support resilient coastal planning and fair adaptation of infrastructure systems.

    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 journalClaims a big stepReal-world use
  3. CerebroSim: Scalable Whole-Brain Simulator at 100-Trillion-Synapse Scale on the LineShine Supercomputer

    Using a model derived from magnetic resonance imaging and diffusion-weighted imaging, CerebroSim simulates 86 billion neurons and 100 trillion synapses on 18,432 nodes across 11.2 million cores of the LineShine Supercomputer, sustaining 24.44 PFlop/s, 91% weak-scaling efficiency, and 94% strong-scaling efficiency.
    AI summary

    The authors created CerebroSim, a framework for simulating an entire brain. It uses three new techniques to handle communication, multithreaded delivery, and memory storage. On the LineShine Supercomputer, it simulated 86 billion neurons and 100 trillion synapses across 18,432 nodes. The system achieved 24.44 PFlop/s with 91% weak-scaling and 94% strong-scaling efficiency. This makes realistic human-brain models practical for studying brain disorders and testing interventions.

    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
  4. Nuclear Genome Amplification of Antibody VDJ Regions in Wastewater

    Here, we report the successful amplification and sequencing of human immunoglobulin heavy and light chain variable regions from wastewater.
    AI summary

    The authors amplified and sequenced human antibody genes from wastewater. They found diverse antibody sequences from different cell lineages. The sequences matched conserved structural regions and variable complementarity-determining regions, or CDRs. The wastewater reads matched known antibody databases at conserved CDRs 1 and 2, but differed at CDR3 sites that control binding. This work shows that wastewater analysis may expand from detecting pathogens to studying population immunity.

    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. Simulation of a Battery Cell on Quantum Computers: Reactions & Transport

    We show how it can evaluate general electrochemical models and present a quantum simulation of the Single Particle Model with electrolyte (SPMe) as the first quantum simulation of a battery cell.
    AI summary

    Simulations of electrochemical materials help technology advance, but computer power limits them. Quantum computing could store far more data, but current machines are too noisy. The authors designed a hybrid quantum-classical algorithm that splits problems into smaller tasks. They show how to implement it for complex equations describing electrochemical systems. They present the first quantum simulation of a battery cell using the SPMe model.

    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
  6. StudentBench: AI and human tutoring yield equivalent GRE learning gains

    We establish that AI tutoring is statistically equivalent to expert human tutoring for GRE learning gains (p = .015), and in five of the seven GRE domains, the best performing AI tutor surpassed the human tutor, on average.
    AI summary

    The researchers created StudentBench, a platform to test whether AI tutors can match human tutors for learning gains. They tested 2,383 people on GRE questions with AI tutoring, human tutoring, or no tutoring. AI tutoring was statistically equivalent to expert human tutoring for overall learning gains (p = .015). In five of seven GRE areas, the best AI tutor beat the human tutor on average. One AI tutor matched human tutoring gains (p = .044) at 918 times lower cost: $0.0052 versus $4.81 per percentage point gained.

    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 useEasy to readCode available
  7. Low-Intensity Focused Ultrasound of the Amygdala in Depression and Anxiety: A First-in-Human Active-Controlled Trial

    ConclusionsAmygdala-targeted FUS produced anatomically specific, multimodal evidence of neuromodulation relative to active control, with a manageable safety profile.
    AI summary

    Researchers tested low-intensity focused ultrasound on the amygdala in ten people with major depression. They compared amygdala treatment to a control treatment on a different brain area. No serious harms occurred, but side effects were more common with amygdala treatment. Brain scans showed the amygdala treatment changed blood flow and activity in specific connected regions. Participants reported feeling calm, clear, or light only after amygdala treatment, not after control.

    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
  8. GUIAuditor: Enabling Post-hoc Child Safety Forensics via Action-Guided GUI Provenance on Mobile Devices

    We present GUIAuditor, the first system designed to realize this vision by creating GUI Provenance: a queryable, semantic record of a child's interaction sequence.
    AI summary

    Current automated tools for protecting children online often fail because they produce false alarms or miss real dangers. They built GUIAuditor, a system that uses a multimodal AI model to turn sequences of screen interactions into readable summaries. Their method cuts data volume by over 89.2% compared to standard techniques, with minimal accuracy loss. In tests, the system correctly identified important events 95.23% of the time and found the right evidence for over 90.20% of natural language questions. The full system runs on modern smartphones, using about 2.1 watts of extra power and 7.4 seconds of delay per event.

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

    Preprint with a published versionClaims a big stepReal-world use
  9. ForgetMimic: Motion Unlearning for Reinforcement Learning Humanoid Control

    To this end, we propose {ForgetMimic}, the first motion-level unlearning method designed specifically for physical-world humanoid control.
    AI summary

    Reinforcement learning lets humanoid robots learn agile movements from human demonstrations. The authors created ForgetMimic, the first method to make robots forget specific motions while keeping others. This matters for safety, privacy, and legal reasons like removing harmful or copyrighted movements. They tested it on two real humanoid robots across twelve motions including dance, fight, and flip. Results show ForgetMimic successfully removes target motions without harming other learned behaviors.

    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 useCode available
  10. Creases gate and steer droplets via elastocapillary repulsion

    Creases, localized indentations arising from elastic instabilities at strained soft interfaces, autonomously gate and steer droplets via long-range elastocapillary repulsion.
    AI summary

    Creases are small dents that form on stretched soft surfaces. They can block or guide droplets without any external control. Small droplets stop before reaching the crease, while large droplets pass through. Slight changes in stretch create large changes in which droplet sizes get blocked. The researchers built filters and programmable fluid devices using this effect.

    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
  11. Stress triggers global histone synthesis in the liver

    However, multiple systemic and liver-specific stresses induce a previously uncharacterized surge of hepatocyte nucleosome production that is uncoupled from DNA replication.
    AI summary

    Researchers developed chemical labeling methods to measure histone synthesis rates in living animals for the first time. Normally, histone synthesis matches tissue growth closely. However, various stresses triggered a sudden increase in nucleosome production in liver cells that was not linked to DNA replication. Newly made nucleosomes were deposited at gene promoters, which was needed to control transcription levels. These findings may help design ways to adjust histone production to affect stress responses or disease.

    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
  12. Generalizable Robotic Insertion with World Models

    Our model-based approach trains a single world model on up to 90 insertion tasks with geometrically diverse parts, achieving 56% zero-shot success on unseen objects with unknown geometry compared to just 7% with a model-free baseline.
    AI summary

    Current robotic assembly systems need separate specialized policies for each insertion task, making new deployments slow and tedious. The authors built a single world model using robot body sensing and raw video from a wrist camera. Trained on up to 90 tasks with diverse parts, it achieved 56% success on never-seen objects versus 7% for a baseline method. Performance improves as more objects are added to training, showing strong scalability. Fine-tuning the general model on new objects uses data more efficiently than training from scratch and sometimes reaches better final performance.

    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. Observation of the orbital Nernst effect

    Here, we report the observation of the orbital Nernst effect in Ti.
    AI summary

    The orbital Nernst effect is a newly predicted phenomenon where heat flow creates a sideways orbital current. They used nickel electrodes to detect a voltage that changes with magnetization direction and grows linearly with temperature difference. This discovery opens the door to a new field called orbital caloritronics.

    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
  14. Transport evidence for the orbital Nernst effect

    Here, we report the first experimental evidence for the orbital Nernst effect by measuring the magneto-thermopower in the yttrium iron garnet (YIG)/Pt/CuO x heterostructures.
    AI summary

    The orbital Nernst effect is a thermal version of the orbital Hall effect, where heat drives a sideways flow of orbital angular momentum. Researchers have now found the first experimental proof of this effect. They measured magneto-thermopower in layered structures of yttrium iron garnet, platinum, and copper oxide. This finding suggests that orbital properties could help harvest heat energy in common, abundant materials.

    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 journalClaims a big step
  15. Chemogenetic activation of hypothalamic oxytocin neurons reorients face exploration toward the eyes in macaques

    These findings provide the first causal evidence in primates that hypothalamic OT neurons regulate social visual orienting by increasing exploration of the eyes, thereby linking a hallmark effect of intranasal OT administration to defined central OT pathways and establishing chemogenetic access to this system in freely behaving macaques.
    AI summary

    The researchers developed a method to switch on specific oxytocin brain cells in macaques using a technique called chemogenetics. They tracked where the animals looked on blurred face images using a finger-controlled window. Activating the oxytocin neurons made the animals spend more time looking at the eyes and less time on other face areas. This is the first causal evidence in primates that these brain cells control social visual behavior by increasing eye exploration.

    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
  16. Orthogonal Chemical Proteomic Strategies Reveal the Cholesteryl Ester Interactome in Mammalian Cells

    Collectively, this work provides the first systems-level map of CE-protein interactions and establishes a generalizable, multimodal framework for chemically resolving lipid-protein interactomes.
    AI summary

    Scientists built a new platform to find which proteins interact with cholesteryl esters in mammalian cells. They used three different photoaffinity labeling methods to tag and capture these interactions. The combined approach identified 495 proteins that associate with cholesteryl esters. The three methods found different sets of proteins, showing that how the lipid is made and delivered affects which proteins it meets. Many of these proteins are linked to diseases and can be targeted by drugs, connecting cholesteryl esters to metabolic, neurological, and cardiovascular conditions.

    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. Longitudinal, Non-Invasive Imaging of the Developing Chick Heart, Vasculature, and Chorioallantoic Membrane (CAM)

    We present an automated laser speckle contrast imaging (LSCI) platform for non-invasive, longitudinal, label-free imaging of the developing heart and extraembryonic vasculature of the chick embryo through the intact eggshell.
    AI summary

    The authors built an automated laser speckle contrast imaging system to watch chick embryo hearts and blood vessels grow without opening the egg. The system can spot blood vessels as small as 70 micrometers and allows repeated imaging from day 0 to day 8 of incubation. It captures both the structure of growing blood vessels and functional data like blood flow and heart rate. Using this system, they made the first continuous time-lapse recordings of the early working chick heart and its extraembryonic vessels. They found slight differences in early heart and vessel development across 15 embryos from different chicken breeds with white, brown and green eggshells.

    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. EvoAudio: Recursive Self-Improvement for Audio Understanding

    We therefore propose EvoAudio, a recursive self-improvement system for audio understanding.
    AI summary

    Audio language models understand speech content better than sound quality. The authors created EvoAudio, a system that improves itself recursively. It is the first system to evolve the model, audio, questions, and difficulty together in a closed loop. EvoAudio uses the current model's performance to set the next training data's focus and difficulty. Across 13 rounds, it improved five different models by up to 6.3 points on audio understanding tests.

    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
  19. First Pre-peak Ultraviolet Spectrum of a Tidal Disruption Event: A Fast Outflow Revealed Prior to Maximum Light in TDE2025aarm

    Our HST/STIS spectrum shows strong evidence for an early-time outflow through broad near-UV (NUV) and far-UV (FUV) absorption lines blueshifted by 10,000 km s, including two new UV broad absorption features not yet identified in a TDE.
    AI summary

    Scientists captured the first ultraviolet spectrum of a star being torn apart by a black hole, taken about 20 days before the event reached peak brightness. The spectrum shows strong evidence of a fast outflow moving at roughly 10,000 kilometers per second, revealed through broad absorption lines in the ultraviolet. The light from this event does not follow a simple blackbody pattern, suggesting the outflow reprocesses the radiation. This finding implies that previous estimates of total brightness for similar events may have been underestimated by about a factor of nine. The study confirms that these tidal disruption events can launch fast outflows very early and highlights the value of early far-ultraviolet observations.

    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
  20. The next Fourier transform: rewiring, representation, and artificial mathematical creativity

    This motivates our Rewiring Thesis: a conceptual revolution is a persistent, large-scale reorganization of transfer relations among mathematical problems.
    AI summary

    They propose that revolutions reorganize how math problems connect, which methods transfer between them, and what new questions open up. Their Rewiring Thesis states that conceptual revolutions are lasting, large-scale reorganizations of these transfer relations. They suggest AI should be tested not just on solving conjectures, but on inventing new conceptual structures like 'the next Fourier transform'.

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

    PreprintBroad interest