Your AI research companion

1. cs.CLContextual Linear Activation Steering of Language ModelsBrandon Hsu, Daniel Beaglehole, Adityanarayanan Radhakrishnan, Mikhail Belkin

   The paper presents a way to control a language model whose strength changes depending on the current text, rather than applying the same adjustment everywhere. This allows it to steer different behaviors at different points in one response while keeping …

2. cs.AIWMIP: A Working-Memory-Inspired Programmable Probe of Cumulative State Tracking in Large Language ModelsDengzhe Hou, Lingyu Jiang, Deng Li, Zirui Li, Fangzhou Lin, Kazunori D Yamada

   The authors create a test of whether language models can keep track of information as it changes repeatedly within a prompt. Across 18 models, performance drops sharply as updates accumulate, but written working steps and answer-format instructions greatly improve scores.

3. cs.AIOn Discovering Algorithms for Adversarial Imitation LearningShashank Reddy Chirra, Jayden Teoh, Praveen Paruchuri, Pradeep Varakantham

   The paper uses an automated search process guided by a language model to discover how imitation-learning systems should turn comparisons with expert behavior into training rewards. The resulting method performs better than human-designed alternatives, works in new settings, and makes …

Every weekday, the new papers in your categories, each with a short summary. See a whole digest →

New preprints explore robots learning from simulation, AI-assisted mathematical discovery, and why accuracy depends on what—and where—we choose to measure.

6 papers put into perspective

1. The coastline looks different outside the training set
2. When more accurate arithmetic breaks a model
3. AI text spreads unevenly across the web
4. Cleaner movement data can change the movement being measured

1. astro-ph.EPPrecision Requirements for Robust Small Planet ClassificationCayman T. Unterborn, Stephen R. Kane

   The authors assessed whether measurements of small planets are precise enough to tell if they are rocky or have other compositions. For many planets, the required improvement in measurements may be so large that studying their atmospheres will also be …

2. astro-ph.EPThe Hot Neptune Initiative (HONEI): IV. TOI-4129 b, an Adolescent Neptune in the RidgeSureena Lakhanpal, Shreyas Vissapragada, Luca Naponiello, Simone Filomeno, Kati…

   The researchers confirmed a young, low-density Neptune-sized planet that orbits its star every 3.27 days, near a group of planets whose origins are not yet understood. Its age and orbit make it useful for testing how planets may move close …

3. astro-ph.HEJets and winds from the most powerful compact object in NGC 5408Roberto Soria, Manfred W. Pakull, Christian Motch, Thomas D. Russell, Fabien Gr…

   The authors study an unusual source in the galaxy NGC 5408 using radio, optical, and X-ray observations. They argue that it is a powerful stellar-mass black hole feeding rapidly, with gas and possibly a jet obscuring much of its X-ray …

Every weekday, the new papers in your categories, each with a short summary. See a whole digest →

This week’s preprints uncover an overlooked outer Solar System object, revisit a stubborn chemical puzzle and use light from stellar remnants to reconstruct what cannot be seen directly.

6 papers put into perspective

1. A bright world in a neglected patch of sky
2. A chemical discrepancy becomes a temperature problem
3. Measuring a neutron star across the spectrum
4. A supernova with a delayed return

1. quant-phHigh-fidelity quantum gate learning via physics-informed neural networksLuis Fabra, Cristian E. Susa, Alcides Montoya, Ariel Norambuena

   The authors train neural networks to design the time-varying controls needed to perform quantum logic gates accurately. Their simulations show very small errors for several gates and examine how the controls behave when noise and practical implementation limits are included.

2. quant-phEntanglement-Aware Quantum Circuit Optimization through Globally Optimal PartitioningGregory Morse, Muhammad Al Farizi, Jakab Nádori, László Hajas, Tamás Kozsik, Pe…

   The authors develop a way to divide quantum circuits into blocks and redesign each block while accounting for how strongly its qubits are linked. In tests, this approach produced more verified circuit designs than a comparison method and never used …

3. quant-phLong-range entangled criticality from coherent errorsAmir-Reza Negari, Subhayan Sahu, Jan Behrends, Benjamin Béri, Timothy H. Hsieh

   The authors show that coherent errors can turn a protected quantum code into a state with long-range quantum correlations. They develop general tests for when such correlations persist and use them to identify a broad critical regime in a surface-code …

Every weekday, the new papers in your categories, each with a short summary. See a whole digest →

New preprints explore how quantum behavior is lost, preserved and engineered—and how a seemingly reversible change can undermine quantum communication.

6 papers put into perspective

1. Watching an electron lose its quantum character
2. For silicon qubits, quieter can also mean harder to drive
3. A small quantum processor hosts a fractional quantum state
4. Repeated measurement slows a spin system’s forgetting

1. cond-mat.mes-hallLiouvillian Kramers DegeneracyKohei Kawabata

   The authors show that a symmetry-related degeneracy can occur in the mathematical description of open quantum systems, even when the systems lose energy or information to their surroundings. They construct examples and show that their spectral patterns belong to a …

2. cond-mat.softStationary distribution of active Brownian particles in a bounded domainAlexandra Tzella, Daniel Loghin, David J. Needham, Jacques Vanneste

   The paper studies how self-propelled particles spread through a confined space and gather at its walls. It explains why the particle distribution becomes sharply concentrated near points where particles leave the wall, and how a small amount of ordinary random …

3. cond-mat.mtrl-sciPossibility of Potts Spin-Ice in Rare Earth Twisted-Kagome Layers: an Example of DyAgGeGunnar F. Schwertfeger, Po-Hao Chang, Predrag Nikolic, Igor I. Mazin, Hari Bhan…

   The authors argue that DyAgGe, a magnetic material with a kagome-like structure, has a kind of spin ice in which each site can point in four directions rather than two. This identifies a new candidate for studying unusual magnetic states …

Every weekday, the new papers in your categories, each with a short summary. See a whole digest →

This week’s selected preprints use interference, ultrafast light, electron correlations and living swimmers to reveal how microscopic behavior becomes measurable collective order.

6 papers put into perspective

1. Distant clocks become a probe of gravity
2. A superconducting state reveals its internal sign pattern
3. Light opens a topological gap in graphene
4. Clockwise swimmers build a counterclockwise fluid

1. gr-qcReference-renormalized Jacobi primitives for finite-distance massive-particle deflectionReggie C. Pantig

   The authors propose a method to calculate how massive particles bend near spherical objects when the source and observer are at finite distances. Unlike earlier versions, it does not require a suitable circular orbit to set a reference, so it …

2. gr-qcGauge-invariant non-universal scalar couplings in teleparallel gravityGiovanni Otalora, Rudranil Roy

   The authors formulate a way for a scalar field’s gravitational coupling to vary while preserving the theory’s gauge symmetry. In example cosmologies, they find conditions under which this variation fades over time, allowing expansion patterns that approach power-law inflation or …

3. hep-thLinear Response in the $\mathcal{PT}$-Broken Phase: Conductivity of Fermions with Complex Conjugate MassesV. E. R. Lemes, D. G. Tedesco

   The authors study how a model of fermions with complex masses responds to an applied signal, and find that its two response channels behave very differently. One channel cannot have a positive-probability interpretation, while the other has a positive spectrum …

Every weekday, the new papers in your categories, each with a short summary. See a whole digest →

New preprints use neutrinos, nuclear decay, colliding nuclei, numerical QCD and gravitational waves to test how well our descriptions of nature survive demanding measurements.

6 papers put into perspective

1. A clearer signal from reactor neutrinos
2. Small nuclear collisions show a collective pattern
3. An exotic nucleus tests how particles decay together
4. Computing the thermodynamics of dense quark matter

1. bioRxiv: NeuroscienceLAMA2-related muscular dystrophy alters Schwann cell calcium homeostasis and increases peripheral nerve infla…J. Wang, M. Khati, M. Heron, A. Munir, O. Akinyele, M. A. Johnson et al.

   The researchers examined peripheral nerves in a mouse model of LAMA2-related muscular dystrophy from early life into adulthood. They found lasting nerve and myelin abnormalities alongside changes in Schwann-cell calcium and increased immune-cell presence, pointing to processes that may contribute …

2. bioRxiv: NeuroscienceSex differences in fear extinction are conditional, not categoricalM. T. Nguyen, A. Didachos, B. Roozendaal, M. J. A. G. Henckens, K. Gulmez Karaca

   The researchers compared how male and female mice learn to reduce fear after a threatening experience, including whether training length and testing time matter. The sexes behaved similarly in most tests, with a difference in fear reduction appearing only during …

3. bioRxiv: NeuroscienceEigenBrain maps cortical molecular architecture from functional, geometric, and structural eigenmodesB. Wan, P. Du, A. Ferrari, A. Saberi, A. John, J. Yan et al.

   The researchers tested whether patterns in MRI data can help estimate the brain’s molecular maps when direct molecular scans are unavailable. Functional MRI patterns could approximate eight receptor and transporter maps, and the approach also detected group differences in autism …

Every weekday, the new papers in your categories, each with a short summary. See a whole digest →

A simulated fly, a stress test for seizure prediction and studies of developing brains and inherited risk reveal why biological explanations depend on what surrounds the system being studied.

5 papers put into perspective

1. A fly connectome takes its first steps
2. The seizure predictor depends on the test
3. Seeing helps blood vessels mature
4. Birth changes the developmental clock

1. bioRxiv: Cell BiologyPathological levels of α-synuclein increase basal mitophagy in Parkinson's dopamine neuronsE. D. Mock, B. Vallin, W. McGuinness, M. C. Caiazza, M. Lavelle, B. Jenkins et …

   The study examined dopamine neurons carrying an extra copy of the alpha-synuclein gene, a cause of early-onset Parkinson’s disease. These neurons broke down mitochondria too actively, and blocking a process that splits mitochondria improved several measures of cell function.

2. bioRxiv: Cell BiologySpatiotemporal mapping of endothelial proliferative response in acute chemical liver injuryA. S. Rajput, H. Yoon, K. L. Travis, E. J. Hernandez, E. P. Shah, D. B. Azizoglu

   The researchers mapped when and where blood-vessel cells in the liver divide after chemical injury. The cells divide later than liver cells and mostly near the center of liver lobules, helping explain how the liver’s blood vessels respond during repair.

3. q-bio.BMAblating Design Choices for Protein Ensemble Generators Across Structural RegimesEoin Quinn, Oliver Bent

   The researchers tested which design choices matter most in models that generate ensembles of protein shapes, comparing their outputs with experimental measurements for different kinds of proteins. The source of the training data mattered more than several architectural choices, and …

Every weekday, the new papers in your categories, each with a short summary. See a whole digest →

This week’s preprints trace how nerves, immune cells, fungi and membrane proteins shape disease—and how a diagnosis can remain absent even when the laboratory evidence is present.

6 papers put into perspective

1. A nerve reflex opens the lung’s vascular barrier
2. An adjustable doorway for folded proteins
3. Engineered immune cells need reinforcements
4. A hemorrhagic fever virus depends on host entry receptors

1. physics.geo-phShallow lunar fractures dominate the long-lasting seismic codaLei Zhang, Qian Huang, Minghao Huang, Jinhai Zhang, Dijun Guo

   The authors propose that cracks in the Moon’s shallow crust cause the long-lasting reverberations seen in lunar earthquake records. Their simulations reproduced Apollo seismic signals more closely than earlier models and suggest the uppermost kilometer is especially important.

2. physics.geo-phExtracting and Characterizing Near-Fault Effects from Strong Motion Observations of the 2022 Chishang Earthqu…Grigorios Lavrentiadis, Domniki Asimaki, Trevor J. Carey, Henry Benjamin Mason

   The researchers developed a way to process earthquake recordings that preserves lasting ground shifts and strong bursts of motion near a fault. Applied to recordings from two 2022 Taiwan earthquakes, the method agreed with satellite-based measurements and revealed patterns linked …

3. astro-ph.EPPrecision Requirements for Robust Small Planet ClassificationCayman T. Unterborn, Stephen R. Kane

   The authors assessed whether measurements of small planets are precise enough to tell if they are rocky or have other compositions. For many planets, the required improvement in measurements may be so large that studying their atmospheres will also be …

Every weekday, the new papers in your categories, each with a short summary. See a whole digest →

New preprints examine what forest treatments protect, how parasites silence collective defenses, and what clocks, comet dust and weather models can reveal about a changing world.

5 papers put into perspective

1. Saving trees without restoring a fire-adapted forest
2. A parasitoid cuts the alarm line
3. Interstellar dust with little sign of heating
4. Clocks follow the changing pull of the tides

1. math.PRMarkovian Lifts and Multifactor Approximations for Weighted Sub-fractional Brownian MotionJ. H. Ramirez-Gonzalez, Edison Cuba

   The paper gives a way to represent weighted sub-fractional Brownian motion as a Markov process and approximate it using a collection of simpler random processes. It proves convergence and gives error rates, supporting practical simulations of equations driven by this …

2. math.PRRényi mixing bounds for kinetic samplers via $L^2$ hypocoercivityJianfeng Lu, Yinchen Luo

   The paper studies how quickly three methods for sampling from complex probability distributions approach their target distributions. It proves faster bounds under stated assumptions and shows that some of these bounds cannot generally be improved.

3. math.OCExact penalty methods with adaptive discretization for nonsmooth semi-infinite optimizationGiancarlo Bigi, Riccardo Tomassini

   The paper develops penalty-based algorithms for optimization problems with infinitely many constraints and nonsmooth functions. It proves conditions under which the methods preserve useful stationary behavior and reports numerical comparisons with existing approaches.

Every weekday, the new papers in your categories, each with a short summary. See a whole digest →

New preprints revisit black-hole thermodynamics, geometric efficiency, numerical algorithms and the limits of what simple models can represent.

6 papers put into perspective

1. An extremal black hole, reached through vacuum
2. The geometry of enclosing space
3. A speed limit for finding roots
4. A community can fail despite reassuring interaction signs

1. cs.CRLoRA-Patch: Defending Against Backdoored LoRA Adapters via Directional SteeringShengyan Huo, Peihai Jiang, Xixiang Lyu, Yige Li

   The paper introduces a way to reduce hidden malicious behavior in add-on components used to adapt language models, while leaving the original model unchanged. Tests across several attacks and models found that it reduced trigger-based behavior while preserving normal task …

2. cs.CRFixed Detectors, Moving Code: How Well Do Static Security Analyzers Hold Up over a Decade of C/C++ Evolution?Junhua Tang, Zhengsong Zhang. Haipeng Cai

   The paper tests whether software security scanners keep working well as C and C++ code changes over time. Across thousands of vulnerable and fixed cases, it finds that apparent improvements or declines can reflect broader alerts, changes in analysis coverage, …

3. eess.SPMission Capability-Aware UAV Swarm Coordination for Autonomous Disaster ResponseMudassar Shakeel, Taha Mahmood, Ahmed Naeem, Muhammad Hamza Jehangir, El Mehdi …

   The paper proposes a system for assigning rescue drones to disaster areas based on what each drone can do and what each area needs. Simulations indicate that adjusting assignments and routes in this way improves how well the swarm meets …

Every weekday, the new papers in your categories, each with a short summary. See a whole digest →

Among this week’s computing preprints, progress comes from reorganizing calculations, fitting speech recognition onto a microcontroller, and testing whether elaborate systems actually earn their complexity.

6 papers put into perspective

1. A famous coloring problem gets a faster solution
2. Counting a disordered system up to a critical boundary
3. Speech recognition without a neural accelerator
4. A newer GPU is not automatically a better purchase

1. econ.GNAgreements Have a Clock: Compute Growth and the Verification of AI RestraintLi Gan

   The paper models what makes an agreement to pause frontier AI development possible and how monitoring could enforce it. It finds that growing computing power shortens the time available to respond to secret advances, while matched safety research and cheaper …

2. econ.GNPrizes, Shared Hazard, and the Speed of Frontier AILi Gan

   This paper models why AI companies may race to develop powerful systems even when faster development raises shared catastrophe risks. In its example, removing competition or coordinating research limits slows the race, while safety audits alone do not.

3. q-fin.STAre LLM Research Agents Reliable Judges of Predictability? Evidence from Planted-Signal WorldsChang Liu

   The authors test whether AI research agents can tell when a forecasting model truly finds a useful pattern, using simulated data where the answer is known. The agents often claim success when no pattern exists, so their judgments need independent …

Every weekday, the new papers in your categories, each with a short summary. See a whole digest →

New preprints examine maternity leave, agriculture’s origins, policy evaluation, high-speed trading and bank capital—and show why the route from data to conclusions matters.

5 papers put into perspective

1. Paid maternity leave and the limits of a policy cutoff
2. Why farming began—and why it spread
3. A feared statistical problem meets a large empirical check
4. What trading speed buys in China