PreprintNot peer reviewedArtificial IntelligencePreprint proposes adaptive routing to cut oracle feedback costs in online RLHFOriginal title: AERO: Adaptive Ensemble-Disagreement Routing for Oracle Feedback in Sample-Efficient Online RLHFA preprint reports a method that targets costly human feedback only where a model's reward estimates are uncertain, cutting query needs.Shrey Singhal et al. · Added 12 Nov 2026
PreprintNot peer reviewedQuantum ComputingPreprint benchmarks 1,368 quantum neural network setups for forecasting exchange ratesOriginal title: Large-Scale Benchmarking of Quantum Neural Network Configurations for Financial Time Series ForecastingA preprint reports a large sweep of quantum neural network designs for GBP/USD forecasting, with hardware noise flagged as a key barrier.Jack Waller et al. · Added 12 Nov 2026
PreprintNot peer reviewedArtificial IntelligenceA preprint reports a code watermarking scheme resilient to LLM rewriting attacksOriginal title: RoCoMark: Robust Code Watermarking Against LLM-Based Functionality-Preserving Rewriting AttacksA preprint describes RoCoMark, a watermarking method trained to survive code rewriting by large language models.Hongjie Zhang et al. · Added 12 Nov 2026
PreprintNot peer reviewedFrom QruiseQuantum ComputingPreprint describes a toolset for modelling and optimising neutral-atom and ion qubit gatesOriginal title: Differentiable Digital Twins for Optimization, Calibration, and Characterization of Neutral-Atom- and Ion-based Quantum ComputersA preprint presents software combining optimal control and automatic differentiation to improve atomic qubit gate fidelities.Added 12 Nov 2026
PreprintNot peer reviewedFrom ByteDanceArtificial IntelligenceResearchers propose scoring text-to-image models by how well prompts can be reconstructedOriginal title: Read It Back: Pretrained MLLMs Are Zero-Shot Reward Models for Text-to-Image GenerationA preprint reports a training-free reward method that uses image-to-text recall to guide reinforcement learning for image generation.Added 12 Nov 2026
PreprintNot peer reviewedFrom Meituan / Meta AIArtificial IntelligencePreprint proposes teacher-guided method to fix flawed reasoning distillation in LLMsOriginal title: Teacher-Guided Policy Optimization for On-Policy Reasoning Distillation under Large Policy DivergenceA preprint describes a new on-policy distillation method that lets a teacher model guide student LLMs even when their outputs diverge widely.Added 12 Nov 2026
PreprintNot peer reviewedFrom Jiffy LabsArtificial IntelligenceResearchers map 284 attack types and gaps in defences for agentic AI systemsOriginal title: Agents of Chaos: A Systematization of Knowledge (SoK) on Malicious Attacks on Agentic AI SystemsA preprint systematically catalogues attacks and safeguards for agentic AI systems, finding uneven and often unverified defence coverage.Added 12 Nov 2026
PreprintNot peer reviewedFrom Northrop GrummanQuantum ComputingResearchers report soldering soliton-based DAC chip to control transmon qubits in nanosecondsOriginal title: Transmon Phase Gates Controlled by Superconducting Soliton DACA preprint describes a superconducting DAC that drives fast transmon gates but introduces a small non-local phase error across many qubits.Added 12 Nov 2026
PreprintNot peer reviewedArtificial IntelligencePreprint tests how well LLMs resist adversarial edits in radiology summariesOriginal title: Adversarial Evaluation of Large Language Models for Patient-Friendly Radiology Report SummarizationA preprint reports that LLMs summarising radiology reports often miss key clinical edits like anatomy or certainty changes.Magdalini Paschali et al. · Added 12 Nov 2026
PreprintNot peer reviewedQuantum ComputingPreprint models how shuttling protocols affect decoherence in multi-spin qubitsOriginal title: Trajectory-Dependent Decoherence and Leakage During Shuttling of Multi-Spin QubitsA preprint studies how different spin-shuttling schemes affect decoherence and leakage in semiconductor logical qubits.Yu-Ning Zhang et al. · Added 12 Nov 2026
PreprintNot peer reviewedFrom HuaweiArtificial IntelligenceResearchers report LLMs struggle to track shifting user needs in multi-turn chatsOriginal title: TurnFlow: Benchmarking User-State Reasoning in Multi-Turn LLM Agents via Scene-Guided User SimulationA preprint introduces TurnFlow, a benchmark finding LLMs falter when user goals or scope change mid-conversation.Added 12 Nov 2026
PreprintNot peer reviewedArtificial IntelligenceA preprint reports pruning method shrinks SO(3) atomistic models without retraining from scratchOriginal title: Compact SO(3) Equivariant Atomistic Foundation Models via Structural PruningA preprint describes structural pruning that compresses equivariant atomistic foundation models while keeping accuracy higher than smaller models trained from scratch.Chen Wang et al. · Added 12 Nov 2026
PreprintNot peer reviewedQuantum ComputingPreprint proposes GKP-encoded protocol for long-range secure direct quantum messagingOriginal title: Long-distance quantum secure direct communication with Gottesman-Kitaev-Preskill qubitsA preprint describes a quantum secure direct communication scheme using GKP qubits that the authors say could work over 1000 km.Meng-Dong Zhu et al. · Added 12 Nov 2026
PreprintNot peer reviewedFrom Meta Platforms / IBMArtificial IntelligencePreprint compares RAG and GraphRAG performance across text-based benchmarksOriginal title: RAG vs. GraphRAG: A Systematic Evaluation and Key InsightsA preprint benchmarks RAG against GraphRAG on question answering and summarisation, finding each has distinct strengths.Added 12 Nov 2026
PreprintNot peer reviewedArtificial IntelligenceNew benchmark finds language models struggle with fine-grained materials science reasoningOriginal title: Benchmarking language models and agents for scientific reasoning in evolving materials laboratoriesA preprint introduces MatScope, a benchmark showing language models handle basic materials calculations well but falter on subtle comparisons.Xuyu Dong et al. · Added 12 Nov 2026
PreprintNot peer reviewedFrom Tata Consultancy Services / IBMQuantum ComputingResearchers report quantum method for larger-scale molecular dynamics simulationsOriginal title: Towards large-scale ab initio Molecular Dynamics Simulation with Efficient Sampling-based Quantum Computing AlgorithmsA preprint describes SqDRIFT-AIMD, a quantum-computing approach said to outperform CCSD for 40-qubit molecular dynamics simulations.Added 11 Nov 2026
PreprintNot peer reviewedQuantum ComputingPreprint compares quantum chip “quantumness” across superconducting and ion-trap hardwareOriginal title: Benchmarking Quantum Computers via Protocols, Comparing Superconducting and Ion-Trap Quantum TechnologyA preprint reports a protocol-based benchmarking method to compare quantumness between superconducting and ion-trap quantum chips.Nitay Mayo et al. · Added 11 Nov 2026
PreprintNot peer reviewedFrom Origin Quantum Computing TechnologyQuantum ComputingPreprint proposes depth-reduced method for simulating open quantum system dynamicsOriginal title: Depth-reduced quantum simulation of Lindbladian dynamics via adaptive variational quantum trajectory compressionA preprint describes a variational technique to compress quantum trajectory circuits for simulating dissipative quantum systems on near-term hardware.Added 11 Nov 2026
Conference paperReview status unknownFrom Visa / Palo Alto NetworksArtificial IntelligenceResearchers catalogue 24 confirmed security flaws in the Model Context Protocol ecosystemOriginal title: Tracing MCP Security Vulnerabilities: Constructing and Analyzing an MCP Security Vulnerability DatasetA study builds a curated dataset of MCP security vulnerabilities from CVE disclosures and analyses their patterns.Added 11 Nov 2026 · Released May 2026
PreprintNot peer reviewedArtificial IntelligenceA preprint links MLLMs’ weak visual reasoning to a reasoning-access gap, proposes a fixOriginal title: Seeing Isn't Enough: Unlocking Visual Reasoning by Bridging the Reasoning-Access Gap with TIGERA preprint argues multimodal models can read images fine but fail to reason on them, and tests a training method to close that gap.Gregory Kang Ruey Lau et al. · Added 11 Nov 2026
