The Morning Preprint

Wednesday 14 October

Showing the most recent batch — Wednesday 14 October. No papers have been ingested today yet.

8455 papers ingested by the estate's stations that day — about 10 minutes to read. The counts and the chart are the estate's own ingest, not arXiv's submission figures.

The 60-second version

Five things from this day's batch, best first — everything below the fold expands on them.

  1. Black hole energy swallowing is much less than expected

    Ultra-compact X-ray binaries show black holes swallow significantly less accreting energy than neutron stars.

  2. AI search finds transferable nucleic acid design algorithms

    Agentic program search uncovered algorithms for nucleic acid design through the AutoNovo framework.

  3. DelegateGuard creates boundaries for LLM agent actions

    DelegateGuard converts user authorization into constraints based on purpose and labels to block unauthorized moves.

  4. Robotics planning time reduced by ten point three percent

    The ATOM framework cut planning error in estimating abstract transition costs by ten point three percent for vehicle navigation.

  5. Thermodynamic consistency guides biochemical reaction network parameters

    RNAGenScape uses property-guided manifold Langevin dynamics to guide mRNA sequence generation while enforcing thermodynamic consistency.

Across the stations

AI Radio

1225 papers today

Today's focus on AI Radio centers on refining how large models handle complex reasoning and long contexts, particularly with works like A Comprehensive Review on RNA Subcellular Localization Prediction A Comprehensive Review on RNA Subcellular Localization Prediction which synthesizes existing knowledge. This work connects to AutoNovo: Agentic Program Search Discovers Transferable Nucleic Acid Design Algorithms AutoNovo, suggesting that agentic search can uncover transferable design algorithms for nucleic acids. These papers explore the limits of current LLM capabilities in reasoning and knowledge acquisition.

The remaining submissions contribute to various areas of model refinement. Papers such as Deep probabilistic model synthesis enables unified modeling of whole-brain neural activity across individual subjects Deep probabilistic model synthesis enables unified modeling of whole-brain neural activity across individual subjects and Learning Dynamic Belief Graphs for Theory-of-mind Reasoning Learning Dynamic Belief Graphs for Theory-of-mind Reasoning add depth to how models simulate human cognition. Other submissions, like SciRerankBench: Benchmarking Rerankers Towards Scientific Retrieval-Augmented Generation with LLMs SciRerankBench, focus on grounding retrieval in scientific contexts.

Overall, the day highlights the tension between scaling models and maintaining factual fidelity and robust reasoning capabilities.

The day on AI Radio

Astrophysics Radio

220 papers today

Today's Astrophysics Radio work focuses on probing black hole event horizons through X-ray binaries, specifically using ultra-compact systems where gravitational radiation dictates accretion rates. The study of the radiative efficiencies in these systems suggests that black holes swallow a significantly larger fraction of accreting energy compared to neutron stars Observational evidence of black hole event horizon from ultra-compact X-ray binaries.

Other papers contribute diverse insights across various areas. Some research explores the dynamics of icy satellites, looking at how eccentricity tides interact with convection in numerical simulations Interaction of eccentricity tide and convection, while others examine cosmological models for dark matter structure formation using zoom-in simulations like COZMIC COZMIC. I. Cosmological Zoom-in Simulations with Initial Conditions Beyond Cold Dark Matter.

The day leaves open the question of how accurately we can constrain cosmological parameters when dealing with complex astrophysical sources, as hinted by work testing the cosmic distance duality relation Directional Tests of the Cosmic Distance Duality Relation using Pantheon+ and BAO.

The day on Astrophysics Radio

Genomics Radio

44 papers today

The day on Genomics Radio focused heavily on parameter optimization within biochemical reaction networks, particularly how to enforce thermodynamic consistency during Bayesian inference. This work specifically addresses the need for accurate parameter estimation in these complex systems by ensuring that the inferred parameters maintain thermodynamic consistency across different experimental setups, which is a crucial step for reliable modeling.

Other submissions contributed various facets of model building and analysis. We saw efforts in developing computational frameworks like EPITIME for integral epidemic models, alongside work on structural connectivity reconstruction using pairwise information flow to overcome the curse of dimensionality. Additionally, papers explored learning when visual context is important for mouse behavior analysis and methods for predicting cognitive tasks from single-channel EEG, showing diverse applications in modeling different biological data types.

Overall, the day presented a mix of core methodological improvements and specific application studies. The work by EPITIME helps structure epidemic modeling, while Characterizing Fungal Infections in the All of Us Research Program provides valuable, albeit imperfect, real-world data for public health concerns. The open question remains how to seamlessly integrate rigorous thermodynamic constraints with the probabilistic nature of inference when modeling dynamic biological systems like reaction networks and epidemics simultaneously.

The day on Genomics Radio

Quantum Radio

442 papers today

Today's work centers on controlling motion in two-ion quantum devices, specifically exploring how to drive one ion as a classical piston using Coulomb interaction while precisely modulating the other's trapping potential Piston control in a two-ion quantum device. This research aims to find a narrow quantum regime that bridges classical and broad regimes and provides methods to control the piston ion's motion, which is key for microscopic quantum devices.

The remaining papers contribute different avenues of study; some investigate the limits of non-Hermitian quantum dynamics showing that scalable advantages are unlikely, while others look at how disorder affects conductance measurements to extract impurity information Disorder information from conductance: a quantum inverse problem. Other works focus on characterizing entanglement through separability criteria and the geometric structure of mixed states Separability criteria and lower bounds of entanglement via realignment, and how different quantum systems behave under phase transitions Phase Transitions in the Classical Simulability of Open Quantum Systems.

Overall, the day's collection touches on fundamental control schemes like piston dynamics Piston control in a two-ion quantum device and theoretical constraints on quantum computation Computational Complexity and Simulability of Non-Hermitian Quantum Dynamics. The central question left open is whether the proposed piston control scheme can be effectively implemented given the inherent complexity suggested by the computational studies in non-Hermitian systems.

The day on Quantum Radio

Robotics Radio

458 papers today

Today's work centers on efficient navigation in complex 3D environments using hierarchical planning, especially as shown by ATOM which reduces planning time significantly through experience-informed refinement. This framework allows for reusable policies across transitions within a scene graph structure, proving effective on an F1Tenth vehicle.

The remaining papers introduce diverse methods to handle different challenges; some focus on improving robot control policies using flow matching gradients in Flow Policy Gradients for Robot Control, while others explore safety through constraint satisfaction like SQ-CBF and the runtime safety certificates in Controller-Agnostic Runtime Safety Certificates for Tool-Using AI Agents.

Ultimately, the day shows progress in combining planning efficiency with robust control; however, a key question remains about how to reliably generalize world models when faced with incomplete information or when needing to adapt policies across different embodied tasks.

The day on Robotics Radio

Security Radio

139 papers today

Security Radio today is focused on how purpose-bound authorization can create a precise boundary for delegated LLM agent workflows. The work called DelegateGuard DelegateGuard shows how to mediate every data move outside the LLM, blocking exfiltration and authority amplification by converting user authorization into capabilities over purpose and labels.

Other papers contribute to this understanding by exploring various attack vectors and defenses. Watermarking Continuous Tabular Data Generation Watermarking Continuous Tabular Data Generation introduces a statistical method for embedding watermarks in tabular data, while Jailbreak Foundry Jailbreak Foundry creates runnable modules from jailbreak papers for standardized benchmarking.

Ultimately, the day leaves open the precise boundary where attenuation-based authorization suffices for delegated agents. We are left wondering how this purpose-bound runtime authorization integrates with the findings from other works like those exploring adversarial robustness or prompt engineering against agent defenses.

The day on Security Radio

The day, in brief

AI Radio focused on using Graph Neural Networks to control biological networks via the GATTACA Framework for managing complex biological systems. Astrophysics Radio found that the radiative efficiency for a black hole is about two to three orders of magnitude lower than that for neutron star systems by combining accretion rates and luminosities. Genomics Radio developed RNAGenScape which uses property-guided manifold Langevin dynamics to guide the generation of mRNA sequences. Quantum Radio demonstrated how time-frequency cat and Gottesman-Kitaev-Preskill states can act as an internal temporal ruler through fidelity revivals in quantum teleportation with single photons. Robotics Radio showed that the ATOM framework reduced the error in estimating abstract transition costs by ten point three percent during repeated navigation tasks. Security Radio introduced DelegateGuard which converts user authorization into constraints based on purpose, labels, operations, sinks, and delegation depth to block unauthorized actions.

The week in the estate's ingest

Seven days of papers the estate ingested. Weekends are striped — the weekend is visible in the data.

1046Thu
1118Sat
893Mon
8455Wed

Show the answer

8455 papers were ingested on Wednesday 14 October.

8455 papers, three ways

Where today's batch landed

By the paper's primary arXiv category.
cs.LG2956
cs.AI1869
cs.CV521
cs.RO364
cs.CL363
quant-ph271
cs.SE147
cs.CR139

The index

The day's strongest 60, by score. Every title links its paper here and on arXiv.

  1. Early Signatures of Memorization in Diffusion Models via Basin Geometry and Cyclic Denoising [arXiv] [episode]
  2. TACO: Ternary Absolute-max Column-wise One-sparse Optimizer for LLM Fine-Tuning [arXiv] [episode]
  3. From a Large-Scale Teacher to Efficient Multimodal Students for Search Relevance Prediction [arXiv] [episode]
  4. Counting with Attention and LayerNorm: Norm Scaling and Representation Geometry [arXiv] [episode]
  5. Learning to Design Convex Relaxations for Nonconvex Optimization [arXiv] [episode]
  6. Neuromorphic Decoding and Dynamical Image Reconstruction from Populations of Spiking Retinal Ganglion Cells [arXiv] [episode]
  7. Solvating machine-learning interatomic potentials for solution-phase chemistry [arXiv] [episode]
  8. Calibrating LLM Explanations to Human Decisions with Reinforcement Learning [arXiv] [episode]
  9. HarnessTTT: Learn to Optimize Harness at Test-Time for Scientific Discovery [arXiv] [episode]
  10. Robustly Reliable Learning for Regression [arXiv] [episode]
  11. Variable Input Masked Autoencoders for Wrist-Worn Accelerometry [arXiv] [episode]
  12. Monarch-Muon: Structure-Aware Muon for Fast and Memory-Efficient Pretraining [arXiv] [episode]
  13. DC-Flow: Target-Conditioned Cyclic Peptide Design via Delayed-Coupling Flow Matching [arXiv] [episode]
  14. GA-SVD: Generation-aware Singular Value Decomposition for LLM Compression [arXiv] [episode]
  15. Split Conformal Interval Length under Huber Contamination [arXiv] [episode]
  16. Beyond Aggregate Drift: Optimal Client-Specific Windows in Federated Learning [arXiv] [episode]
  17. ProML: Progressive Multi-label Learning [arXiv] [episode]
  18. DRanker: A Transformer-Based Multi-Task Ranking Model for the Disney+ Homepage [arXiv] [episode]
  19. Two Linear Programs are Better than One [arXiv] [episode]
  20. Harmonic Representations in Symbolic Music Transformers are Localized and Controllable [arXiv] [episode]
  21. Triage-PiEvo: Evidence Admission for Principle-Evolving Scientific Agents [arXiv] [episode]
  22. Rethinking Opinion Measurement Interfaces for Human-in-the-Loop Optimization [arXiv] [episode]
  23. Disentangling Marginal Capacity from Geometry in Optimal-Transport Label Transfer [arXiv] [episode]
  24. Blending experimental electron density information to protein-ligand interaction [arXiv] [episode]
  25. Zeroth-Order Fine-Tuning of LLMs in Learnable Low-Rank Subspaces [arXiv] [episode]
  26. Cross-Behavior Policy Optimization: Unified Learning from Offline Demonstrations and On- and Off-Policy Rollouts via a Shared Regression Target [arXiv] [episode]
  27. Robust Langevin Sampling under State-Dependent Heavy-Tailed Noise with Applications to LLM Fine-Tuning [arXiv] [episode]
  28. Monarch-MoE: Structured Experts for Multimodal Continual Instruction Tuning [arXiv] [episode]
  29. State-OPD: State-Grounded On-Policy Distillation for Multi-Turn Agents [arXiv] [episode]
  30. CopRoute: Dependence-Aware Conformal Routing for Language Models [arXiv] [episode]
  31. AgentTune: A Unified Training Lifecycle for Tool-Using LLM Agents [arXiv] [episode]
  32. Lower Bounds for Randomized Linear-Oracle Optimization over Strongly Convex Sets [arXiv] [episode]
  33. VQIL: Visible Quantum Interference Learning for Perturbation Response Ranking on Graphs [arXiv] [episode]
  34. SeoulMMOD: A Large-Scale Multimodal Origin-Destination Flow Benchmark [arXiv] [episode]
  35. Inference-Time Steering Across Scaffolds and Epitopes for Sample-Efficient De Novo Nanobody Design [arXiv] [episode]
  36. Sparse-TD-JEPA: Improving Zero-Shot Reinforcement Learning With Sparse Latent Task Representations [arXiv] [episode]
  37. Multimodal Graph Retrieval-Augmented Sequential Recommendation via Collaborative Filtering Paths [arXiv] [episode]
  38. Memorization and Malign Generalization in Conditional Diffusion Models with Random Features [arXiv] [episode]
  39. Sample-Efficient Generative Conformal Prediction [arXiv] [episode]
  40. Evi-VN: Hard Region Guided Virtual Node Evidence Injection for GNN-Based Fraud Detection [arXiv] [episode]
  41. Compile the Table: Query-Calibrated Operator Compression for Tabular In-Context Learning [arXiv] [episode]
  42. PRAXIS: Learning Dynamics of Self-Improving Models with Symbolic Archives [arXiv] [episode]
  43. DADP: Dynamic Activity-Dependent Pruning, A Reverse Hebbian-Inspired Structural Pruning Method [arXiv] [episode]
  44. Cost-Aware Mixture-of-Experts Coordination for Model Markets [arXiv] [episode]
  45. SCORE: Spectral Correlation Estimation for Multivariate Gaussians [arXiv] [episode]
  46. MolP2SBench: Do LLMs Really Reason from Molecular Properties to Structure? [arXiv] [episode]
  47. GlucoPRISM: Protocol-Supervised Pretraining for Cross-Cohort Glucose Transfer [arXiv] [episode]
  48. Sample-Adaptive Zeroth-Order Post-Training [arXiv] [episode]
  49. STAPLE: Static Sparse Upcycling of Pretrained Checkpoints via Embedding Modules [arXiv] [episode]
  50. Expert-level detection of low-voltage intervals in EEG as a robust biomarker of neonatal encephalopathy [arXiv] [episode]
  51. Differential Refresh of Materialized Derived State: From a Single-Age Equivalence Limit to an Optimal Per-Partition Allocation, with a Real-System Validation on Materialized Views [arXiv] [episode]
  52. CircuitGate: Logic-Consistent Circuit-Level Functional Modeling for And-Inverter Graphs [arXiv] [episode]
  53. UniCSI Towards a Universal Wi-Fi CSI Encoder for Ubiquitous Human Sensing [arXiv] [episode]
  54. COPC: Coupled Off-Policy Correction for Asynchronous LLM Reinforcement Learning [arXiv] [episode]
  55. Sharp Asymptotic Theory of Maximum Likelihood Estimation for Gaussian Processes with an RBF Kernel [arXiv] [episode]
  56. Training with Missed Targets in Generative Recommendation: Separating Supervision from Probability Competition [arXiv] [episode]
  57. Evaluating Sequence Assembly Strategies for Differentially Private Synthetic Time-Series Forecasting [arXiv] [episode]
  58. A Closed-Loop Non-Asymptotic Convergence Analysis of PPO with Learned Critics and Clipping [arXiv] [episode]
  59. PoreML: A Data-Driven Framework for Learning Multiphase Flow in Porous Media [arXiv] [episode]
  60. Boosting and the Expressive Power of Simple Weak Learners via the gamma-VC Dimension [arXiv] [episode]

Listen

Everything the estate published on 2026-10-14, in the player. Playing from this page is how the edition learns whether anyone did.

Previous editions