PreprintNot peer reviewedQuantum ComputingPreprint sets theoretical limits on QKD error tolerance and rangeOriginal title: Fundamental Limits on QBER and Distance in Quantum Key DistributionA preprint derives maximum bit-error thresholds and distance bounds for quantum key distribution, including deep-space links.Stefano Pirandola · Added 1 Nov 2026
PreprintNot peer reviewedFrom Geely GoertekArtificial IntelligenceA preprint proposes budget-aware precision allocation for LLM KV cache quantizationOriginal title: MultiAlloc: Budget-Aware Precision Allocation for KV Cache QuantizationA new preprint describes MultiAlloc, which splits a fixed KV cache memory budget between layer and token precision based on offline profiling.Added 1 Nov 2026
PreprintNot peer reviewedQuantum ComputingResearchers link two trapped-ion processors to detect and correct quantum errors remotelyOriginal title: Error Correction in a Distributed Quantum ComputerA preprint reports the first demonstration of error detection and correction shared across two separate trapped-ion quantum processors.E. M. Ainley et al. · Added 1 Nov 2026
PreprintNot peer reviewedFrom Bridgewater AIA LabsArtificial IntelligenceBenchmark finds AI data-science agents struggle with causal identification and abstentionOriginal title: CausalDS: Benchmarking Causal Reasoning in Data-Science AgentsA preprint introduces CausalDS, a synthetic benchmark showing agents recover causal graphs well but fail at identification, abstention and uncertainty.Added 1 Nov 2026
PreprintNot peer reviewedArtificial IntelligenceResearchers report a lightweight fix for knowledge conflicts in retrieval-augmented generationOriginal title: SPARK: Adaptive Parametric Knowledge Participation for Faithful Retrieval-Augmented GenerationA preprint describes SPARK, a method that learns to adaptively control how much a language model relies on its internal knowledge versus retrieved text.Ruochang Li et al. · Added 1 Nov 2026
PreprintNot peer reviewedArtificial IntelligenceA preprint describes a distributed MoE redesign cutting cross-GPU communicationOriginal title: Federation of Experts: Communication Efficient Distributed Inference for Large Language ModelsResearchers report an architecture that groups experts and KV heads to shrink communication overhead in distributed LLM inference.Muhammad Shahir Abdurrahman et al. · Added 1 Nov 2026
PreprintNot peer reviewedFrom WorkatoArtificial IntelligenceResearchers report adaptive data scheduling boosts LLM reasoning RL trainingOriginal title: Learning at the Right Pace: Adaptive Data Scheduling Improves LLM Reinforcement LearningA preprint reports that scheduling training data by semantic clusters and policy difficulty improves reinforcement learning for LLM reasoning over uniform sampling.Added 1 Nov 2026
PreprintNot peer reviewedQuantum ComputingNew scheme avoids chip defects on IBM’s heavy-hexagon quantum layoutOriginal title: Avoiding Defects within Heavy-Hexagon ArchitectureResearchers report a faster defect-avoidance scheme for heavy-hexagon quantum chips, cutting logical error rates on IBM hardware.Hansol Kim et al. · Added 1 Nov 2026
PreprintNot peer reviewedFrom SiemensArtificial IntelligenceResearchers propose ontology-grounded, reasoner-checked benchmarks to test LLM logicOriginal title: Ontology-Grounded, Reasoner-Verified Benchmarks for Evaluating LLM Reasoning in Scientific AIA preprint presents a pipeline that builds multiple-choice reasoning tests from OWL ontologies, with wrong answers verified false by a reasoner.Added 1 Nov 2026
PreprintNot peer reviewedArtificial IntelligencePreprint proposes comparing tool choices before acting to guide long-horizon LLM agentsOriginal title: Choosing Before Acting: Comparative Value Estimation for Long-Horizon Tool-Use AgentsA preprint reports a trained value-comparison model that helps agents pick better next tool calls, improving task success.Yu Li et al. · Added 1 Nov 2026
PreprintNot peer reviewedArtificial IntelligenceResearchers propose Dev-Primitives to give code components agent-like reasoning in software tasksOriginal title: Harness Engineering for Software Engineering via Modular Executable Dev-PrimitivesA preprint introduces HERMES, a harness wrapping repository components with resident LLMs to improve software engineering agents.Haibo Jin et al. · Added 1 Nov 2026
PreprintNot peer reviewedFrom Google DeepMindArtificial IntelligenceStudy links fine-tuning behaviour spread to a model’s ‘default persona’Original title: The Persona Hierarchy Model: Understanding Contextual Generalization in Fine-Tuning LLMsA preprint proposes that fine-tuned behaviours spread widely when training nudges a model's shared default persona rather than a context-specific one.Added 1 Nov 2026
PreprintNot peer reviewedArtificial IntelligenceA preprint reports 2-bit KV cache compression for looped transformersOriginal title: ResidualQuant: KV Cache Quantization for Looped Transformers with 2-Bit ResidualsResearchers report a quantisation method that shrinks KV cache memory in looped transformers while keeping accuracy close to full precision.Heejun Kim et al. · Added 1 Nov 2026
PreprintNot peer reviewedArtificial IntelligenceResearchers trace LLM forgetting to output embeddings of rare tokensOriginal title: A Deafening Silence: Catastrophic Forgetting Lives in the Output Embeddings of Tokens the Data Never SpeaksA preprint reports that catastrophic forgetting during fine-tuning concentrates in output embeddings of under-exposed tokens, and a simple optimizer tweak reduces much of it.Jonghyun Han et al. · Added 1 Nov 2026
PreprintNot peer reviewedFrom Kuaishou TechnologyArtificial IntelligenceResearchers report a reward model that rivals test-based code RL trainingOriginal title: CodeScaler: Scaling Code LLM Training and Test-Time Inference via Reward ModelsA preprint describes CodeScaler, a reward model that matches or beats execution-based reinforcement learning for code generation while cutting inference latency.Added 1 Nov 2026
PreprintNot peer reviewedArtificial IntelligencePreprint reports strategy-guided RL framework boosts long-horizon agent performanceOriginal title: StraTA: Incentivizing Agentic Reinforcement Learning with Strategic Trajectory AbstractionA preprint introduces StraTA, which adds an explicit trajectory-level strategy to agentic reinforcement learning, reportedly improving efficiency and task success.Xiangyuan Xue et al. · Added 1 Nov 2026
PreprintNot peer reviewedFrom Google DeepMindArtificial IntelligenceA preprint presents a unified benchmark comparing RL and foundation-model agentsOriginal title: Agentick: A Unified Benchmark for General Sequential Decision-Making AgentsResearchers report a 37-task benchmark showing no single agent paradigm dominates sequential decision-making, with reasoning harnesses giving large gains.Added 1 Nov 2026
PreprintNot peer reviewedArtificial IntelligenceResearchers propose a stepwise reward derived from a model’s own likelihood signalsOriginal title: Stepwise Intrinsic Rewards for Reasoning in Large Language ModelsA preprint reports a self-generated process reward that credits reasoning steps without human-labelled annotations or auxiliary models.Xiangwei Wang et al. · Added 1 Nov 2026
PreprintNot peer reviewedFrom HuaweiArtificial IntelligencePreprint proposes MetaPermit, a meta-attribute policy system for securing AI agent tool callsOriginal title: MetaPermit: Scalable and Auditable Access Control for AI Agents via LLM-Inferred Meta-AttributesA preprint describes MetaPermit, which infers abstract 'meta-attributes' with an LLM and lets a fixed policy decide whether agent tool calls are allowed.Added 1 Nov 2026
PreprintNot peer reviewedArtificial IntelligenceResearchers report new bioengineering benchmark exposes gaps in top AI modelsOriginal title: BioEVAL: A global, multi-institutional benchmark of large language and multimodal models for bioengineeringA multi-institutional preprint finds leading LLMs score well on bioengineering recall but struggle more with experimental reasoning and visual data.Shun Ye et al. · Added 1 Nov 2026
