Every paper checked — nothing hidden
All papers
Every paper checked for brief #4 of 12–19 Oct 2026 — nothing is hidden. Brief papers first.
All17,313In brief24More (on-topic)3,333Not relevant13,754Not assessed87Older352Could not confirm93Blogs with no paper266All blogs284
| Paper | Topic | Relevance | Feedback |
|---|---|---|---|
| Noise-Space Reward Alignment for Few-Step Generative Models The paper proposes Noise Policy Optimization to align few-step generative models with black-box rewards by steering their initial input noise space. Ranked #302New way to see a problemPreprint4/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Optimizes few-step generative models to retain sampling speed without heavy search costs. | |
| Who May Release an Agent's Promise? Standing over Commitments in Multi-Party Chat Studies whether conversational agents correctly enforce authorization rules when users attempt to cancel prior agent commitments to others. Ranked #302New testPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Studies reliability and authorization boundaries to prevent unauthorized override in enterprise deployments. | |
| ActICL: Post-Training for Agentic Multimodal In-Context Learning The paper proposes an agentic framework enabling multimodal models to identify information gaps and retrieve in-context demonstrations iteratively. Ranked #302New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Investigates multimodal in-context learning and multimodal large language model agents. | |
| Knowing Is Not Telling: Identity Ossification in LLM Representations of Trans-border Communities Proposes a diagnostic framework to study how language models compress fluid trans-border ethnic identities into rigid nation-state categories. Ranked #302Surprising resultPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28 | On-topic Introduces a diagnostic framework and benchmark measuring LLM bias and representation patterns. | |
| A Baseline Assessment for the Agentic Web Technology Stack Benchmarks combinations of language models, communication protocols, and agent frameworks on web interaction tasks. Ranked #302New testPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Introduces a framework to assess agentic web stacks, protocols, and multi-agent frameworks. | |
| Judging the Judges: Confidence and Disagreement in Rubrics as Rewards Proposes an ensemble framework that weights judge confidence and masks disagreements when training models with rubrics as rewards. Ranked #303Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Improves alignment reliability by filtering uncertain and disagreeing judge feedback. | |
| VisualWorldBench: A Fine-Grained Multi-Task Benchmark for Evaluating Visual World Knowledge in MLLMs Introduces a hierarchical multi-task benchmark to evaluate fine-grained visual world knowledge in multimodal large language models. Ranked #303New testPreprint5/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Introduces a fine-grained multi-task benchmark specifically for evaluating visual world knowledge in multimodal large language models. | |
| An all-transistor physical-learning transformer can be trained to perform in-context learning This paper designs and simulates an all-transistor physical circuit using sub-threshold MOSFETs to run transformer attention and in-context learning. Ranked #303Surprising resultPreprint3/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:29 | On-topic Explores physical-learning transistor circuits for low-power, efficient attention and in-context learning. | |
| Fear Is Not Always Fear: Variation in Emotion Label Equivalence Across Datasets Tests cross-corpus emotion classification transfer across three datasets, finding significant loss in discrimination and conceptual disagreement on labels. Ranked #303Surprising resultPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28 | On-topic Evaluates cross-corpus dataset alignment and benchmark label equivalence in NLP classifiers. | |
| Rollout-Time Intervention: Unlocking Agentic Reinforcement Learning Beyond Policy Limits Proposes Rollout-Time Intervention to convert failed agent trajectories into successful off-policy training signals during reinforcement learning. Ranked #303New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Enhances reinforcement learning for LLM agents performing interactive reasoning, planning, and tool use. | |
| SCARCE: Emergent Institutions in Long-Horizon Multi-Agent Coding Systems The paper studies how autonomous coding agents spontaneously negotiate and coordinate access to shared compute resources under deadlines. Ranked #304Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Investigates unplanned coordination risks, emergent norms, and safety in autonomous multi-agent environments. | |
| GeoThinker-$F$: Endogenous Predictive World Representations for Fast Spatial Intelligence This paper proposes GeoThinker-F, an efficient architecture that predicts spatial representations directly from multimodal hidden states without dedicated 3D encoders. Ranked #304New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Improves multimodal LLMs by internalizing spatiotemporal perception without external 3D encoders. | |
| Error-Based Bidirectional Predictive Coding Enables Deep, 4-Bit Predictive Coding Networks This paper combines bidirectional and error-based predictive coding to enable deep, 4-bit quantized predictive coding networks. Ranked #304New way to see a problemPreprint3/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:18 | On-topic Enables deep, low-cost predictive coding networks using 4-bit quantization for hardware acceleration. | |
| People Making Conversation: Evaluating How Naturally Speech Models Converse Evaluates whether real-time speech models decide to speak appropriately during natural pauses and turn ends in unscripted conversations. Ranked #304New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28 | On-topic Presents a benchmark and evaluation protocol assessing natural turn-taking dynamics in conversational speech models. | |
| SWE-Critix: Post-Training Generalist Verifiers for Assessing Coding-Agent Trajectories Proposes SWE-Critix, a post-training pipeline designed to train verifiers that evaluate and explain long coding-agent execution trajectories. Ranked #304New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Studies trajectory verification and behavior monitoring for autonomous coding agents. | |
| StepSafe: Learning Timely Intervention with Context-Aware Safety Reasoning The paper introduces a training framework to teach AI agents when to intervene proactively and safely during multi-step tasks. Ranked #305New way to see a problemPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Directly targets real-time safety guardrails and risk intervention in sequential agent workflows. | |
| Spectral World Renderer: Unified 3D Hyperspectral Unmixing The paper introduces a synthetic world generator and benchmark for evaluating 3D hyperspectral material unmixing within novel view rendering. Ranked #305New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Combines 3D scene representation with spectral decomposition enabling text queries of material compositions. | |
| A Dependency-Aware Multi-GPU Training Framework for Temporal Graph Neural Networks This paper presents a multi-GPU training framework for temporal graph neural networks that relaxes event dependencies to reduce synchronization. Ranked #305New way to see a problemPreprint3/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:18 | On-topic Accelerates temporal graph neural network training using partitioned multi-GPU parallelism. | |
| MADS: Multi-Agent Dialog Simulation for Robust Conversational Agents Generates synthetic multi-agent dialogues to benchmark conversational agent robustness against non-cooperative and adversarial user behaviors. Ranked #305New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28 | On-topic Introduces MADS-Bench to evaluate agent robustness across failure-prone conversational trajectories. | |
| Agent4Spec: An LLM-based Agentic System to Optimize API Specifications through Automated Validation and Market Analysis Presents Agent4Spec, a multi-agent framework that automatically validates OpenAPI syntax errors and identifies functional API gaps through market analysis. Ranked #305New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Presents an autonomous multi-agent framework orchestrating validation and market analysis tasks. | |
| Evolving Together, Failing Apart: The Skill-Policy Co-evolution Trap in Self-Improving Agents The paper investigates performance vulnerabilities in self-improving agents when their co-evolved skill banks undergo post-deployment drift. Ranked #306New testPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Assesses agent reliability and fragility under post-deployment skill drift. | |
| WinARDet: Compressed Window Auto-Regression for Long-Horizon Vision-Language Grounding The paper proposes compressed window autoregression to prevent repetitive bounding box errors in vision-language object grounding tasks. Ranked #306Surprising resultPreprint5/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Improves autoregressive vision-language models for grounding across dense visual scenes. | |
| Learning to Extrapolate in Depth with Looped Transformers The paper studies depth extrapolation in looped transformers for graph connectivity, identifying finite iterations beyond which test performance degrades. Ranked #306New way to see a problemPreprint3/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:18 | On-topic Analyzes depth extrapolation in looped transformers, which reuse parameters for parameter-efficient inference. | |
| Specialized Decision Models vs. General-Purpose LLMs: Benchmarking Jev Across Knowledge, Reasoning, and Multilingual Tasks The paper benchmarks a specialized multiple-choice decision model against 19 general-purpose LLMs across knowledge, reasoning, and multilingual tasks. Ranked #306Surprising resultPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28 | On-topic Compares model architectures across 13 standard LLM evaluation benchmarks and tasks. | |
| MENTOR: Learning from Failure with On-Policy Pedagogical Intervention Presents MENTOR, a multi-agent framework that intervenes adaptively during failed reasoning rollouts to prevent zero-reward collapse in RLVR. Ranked #306New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Uses a multi-agent framework to diagnose failures in real-time and provide on-policy instructional prompts. | |
| Analog RF Computing for Integrated Data and AI Flows Over Wireless Networks This paper investigates using analog radio frequency computing over wireless networks to deliver low-power neural network inference to resource-constrained edge devices. Ranked #307New way to see a problemPreprint3/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:18 | On-topic Explores analog RF computing to enable low-power neural network inference on edge devices. | |
| Evaluating Model Capabilities After Benchmark Saturation Tests whether transcripts of correctly solved tasks from saturated benchmarks can predict model performance on unseen harder evaluations. Ranked #307New testPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Enables continued assessment of frontier model capabilities essential for AI safety cases and governance. | |
| PhyMetric: Diagnosing Physical Plausibility in Text-to-Video Generation via Scene-Level QA Introduces a diagnostic question-answering benchmark to evaluate whether text-to-video models follow real-world physical laws. Ranked #307New testPreprint5/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Introduces a benchmark using vision-language model judges to diagnose video generation plausibility. | |
| LegalRewardBench: Benchmarking Reward Models for Legal Preference Evaluation Introduces a human-verified benchmark to evaluate how accurately reward models judge language model outputs on legal tasks. Ranked #307New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | On-topic Introduces a benchmark evaluating reward models and LLM judges on legal preferences. | |
| RESCUE: Is This Alarm Worth Acting On? Counterfactual Evaluation of VLA Runtime Monitors Presents RESCUE, a counterfactual simulation framework to evaluate whether runtime failure monitor alarms enable beneficial robot interventions. Ranked #307New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Studies runtime failure monitors and intervention policies for embodied vision-language-action agent execution. | |
| PremiseSQL: Evaluating Premise Critique and Abstention in Text-to-SQL Systems Introduces a benchmark to evaluate whether text-to-SQL systems can detect flawed query premises and abstain from generating invalid SQL. Ranked #308New testPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Addresses reliability and hallucinated execution in real-world database querying systems. | |
| Anvil: Scaling Embodied Spatial Intelligence through Knowledge Recirculation Presents a unified framework with synthetic data generation and benchmarks to train models on embodied spatial reasoning and actions. Ranked #308New testPreprint5/10 suggested | Multimodal models via arxiv · 17 Oct, 06:17 | On-topic Develops multimodal models that ground spatial relations and actions in visual scenes. | |
| Rethinking Prediction Horizons: Decoupling Model Queries from State Updates in Few-Step MeanFlow The paper studies the impact of separating prediction horizons from state update intervals during sampling in generative flow models. Ranked #308Surprising resultPreprint3/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Improves few-step diffusion sampling efficiency at inference time. | |
| Do correct answers have stable latent knowledge support? The paper proposes stability metrics to measure whether correct outputs in language models are backed by stable internal knowledge representations. Ranked #308New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | On-topic Introduces EpiStableBench to evaluate the internal representation stability of knowledge in LLMs. | |
| CLEAT: Learning Value-Guided Clarification for Interactive Agents Co-trains an interactive agent's task policy with a clarification-value controller to optimize when to ask follow-up questions. Ranked #308New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Focuses on training LLM agents to perform interactive clarification to align actions with user goals. | |
| Effectiveness of Hidden-State Probes for Localizing Reasoning Failures Evaluates whether internal hidden-state probes can identify failure modes and pinpoint failing steps in multi-step reasoning traces. Ranked #309Surprising resultPeer-reviewed7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Investigates detecting and localizing internal failures and hallucinated reasoning steps. | |
| Can Visual Generation Expand What Unified Multimodal Models Know? Investigates whether training multimodal models on visual generation tasks transfers accessible factual knowledge into text question answering. Ranked #309New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 17 Oct, 06:17 | On-topic Investigates knowledge transfer between visual generation and text answering within unified multimodal models. | |
| CARD: Closing the Teacher-Student Loop for Adversarial Robustness Distillation Develops a closed-loop distillation pipeline where student feedback refines teacher models to improve adversarial robustness. Ranked #309New way to see a problemPreprint3/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Proposes a cyclic distillation framework to improve compact student model training. | |
| Sanzaru: Benchmarking Protocol Formation under Epistemic & Communication Asymmetry The paper introduces a benchmark evaluating how multi-agent teams develop communication protocols under asymmetric information and restricted communication channels. Ranked #309New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | On-topic Introduces SANZARU, a benchmark for evaluating multi-agent coordination and communication. | |
| Collaborative Learning and Reasoning over Model Slices Develops a collaborative learning framework that splits large language models into smaller heterogeneous slices to mitigate error propagation. Ranked #309Surprising resultPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Explores collaborative reasoning where multiple heterogeneous model slices interact to solve tasks. | |
| When Reasoning Traces Conflict with Evidence: How Answer Configuration Shapes Evidence Following The paper analyzes how answering via a separate model call improves evidence adherence when reasoning traces conflict with provided context. Ranked #310Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Investigates model reliability and adherence to factual evidence over hallucinated or stale reasoning traces. | |
| Probing Negation Understanding in Vision-Language Models With Object Attention The paper analyzes how vision-language models process negation by comparing model attention mechanisms with human eye-tracking data. Ranked #310Surprising resultPreprint5/10 suggested | Multimodal models via arxiv · 17 Oct, 06:17 | On-topic Investigates how vision-language models process negation and visual object attention. | |
| Drift, Then Spread: Learning One-Step Generative Policies for Online Reinforcement Learning The paper proposes a one-step generative reinforcement learning policy that balances reward improvement and entropy without iterative sampling. Ranked #310New way to see a problemPreprint2/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:18 | On-topic Proposes a one-step generative policy to replace costly iterative sampling during RL inference. | |
| Social Bias in Chinese Large Language Models under Social Media Expressions This paper evaluates social bias in Chinese language models across social media expressions, internet slang, and sarcastic packaging. Ranked #310New testPeer-reviewed6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | On-topic Constructs evaluation probes to benchmark social bias under social media expressions. | |
| Competence-Aware Orchestration for LLM-Based Multi-Agent Systems Presents a framework that dynamically adjusts role activation and multi-agent collaboration based on the model backbone's query-specific competence. Ranked #310Surprising resultPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Optimizes multi-agent coordination, role activation, and communication graphs based on model competence. | |
| Beyond Confidence: Rethinking Uncertainty Calibration in Deep Code Models This study evaluates whether confidence scores and calibration methods effectively identify failure cases in deep learning models for code tasks. Ranked #311Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:19 | On-topic Studies uncertainty calibration to prevent silent failures in critical deployment tasks. | |
| ZoomCap: Evidence-Routed Visual Zooming for Recoverable Visual Information Transfer The paper introduces ZoomCap, an adaptive visual zooming and evidence-routing framework to improve fine-grained image captioning supervision. Ranked #311New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 17 Oct, 06:17 | On-topic Presents evidence-routed visual zooming and evaluation for fine-grained vision-language model captioning. | |
| Softmax-Free Attention in Spiking Transformers as the Zero-Temperature Limit of Two-Level Systems This paper provides a thermodynamic framework explaining energy efficiency and optimization issues in softmax-free spiking transformers. Ranked #311New way to see a problemPreprint2/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:18 | On-topic Analyzes energy-efficient, softmax-free spiking transformer attention mechanisms. | |
| The Company a Question Keeps: Auditing Open Jev-Style Models for Retrieval This paper audits request independence in open Jev-style retrieval models across multi-passage query states. Ranked #311Surprising resultPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | On-topic Audits evaluation reliability and behavioral flips in open retrieval checkpoints. | |
| DRYLAB-Bench: Benchmarking the Biosecurity Risk of Agentic LLM Mutation Design Introduces a benchmark evaluating the risks of LLM agents designing dangerous biological mutations when augmented with biodesign tools. Ranked #311New testPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Evaluates tool-augmented LLM agents executing multi-round planning and mutation design workflows. | |
| SoftCache: Adaptive Step Size Caching for Diffusion Transformers Introduces an adaptive step-size caching method for diffusion transformers to speed up generation by reusing intermediate network evaluations. Ranked #312New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Proposes adaptive step size caching to accelerate multimodal diffusion and flow transformers. | |
| Prune the Summary, Not the History: Source-Judged Compaction for Long-Horizon Agents Introduces a method to filter unsupported compressor inferences from long-horizon agent summaries to avoid compounding execution errors. Ranked #312New way to see a problemPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Prevents agent errors and ungrounded hallucinations during context compaction. | |
| PotentialEP: Backprop-Free Training of CIFAR-Scale Generative Models via Equilibrium Propagation The paper introduces a backpropagation-free training method for generative relaxation models suitable for analog neuromorphic hardware execution. Ranked #312New way to see a problemPreprint2/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:18 | On-topic Introduces backpropagation-free training and relaxation sampling for low-power neuromorphic execution. | |
| Words to World: Benchmarking LLMs in Interactive Physical Environments Introduces a closed-loop benchmark testing language models controlling physical robotic hardware and reacting to real sensor feedback. Ranked #312New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | On-topic Introduces an interactive benchmark evaluating models on closed-loop execution in physical environments. | |
| LongHarness: Advancing Agentic Reasoning over 100M-Token Corpora via Stateful Evidence Management Presents an agent harness that coevolves evidence and conclusions to improve reasoning across 100-million-token document corpora. Ranked #312New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Develops an agent harness managing multi-step evidence gathering and iterative conclusion updating. | |
| When Models Misread Ink: Auditing VLMs for Arabic Calligraphy Recognition The paper evaluates vision-language models on Arabic calligraphy recognition and identifies frequent default-class prediction failures. Ranked #313New testPreprint5/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Audits vision-language models on fine-grained visual script recognition. | |
| VoteGEO-Bench: Measuring the Impact of Generative Engine Optimization on Electoral Visibility The paper presents VoteGEO-Bench to measure how rewriting web content influences information visibility in generative search engine election queries. Ranked #313Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Studies manipulation risks and reliability of generative search engines during democratic elections. | |
| ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence The paper introduces a neuro-symbolic benchmark to evaluate whether language model agents actively investigate environments and make causal decisions. Ranked #313New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | On-topic Introduces ACED-BENCH, a new benchmark to evaluate LLM agents on active experimentation and causal decisions. | |
| PreSTO: Predictive Subtree Prefetching for Fast LLM Power Sampling Presents a predictive subtree prefetching method to accelerate power sampling for test-time language model reasoning. Ranked #313Preprint0/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:28 | On-topic Introduces predictive prefetching to speed up sampling and reduce inference latency. | |
| Ecology of AI Agents: Collaboration Creates a Population Threshold for Takeoff Formulates a population dynamics model to study takeoff thresholds and self-replication risks in collaborative AI agents. Ranked #313New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Models the collective takeoff and cyberattack dynamics of collaborating AI agent populations. |
1,861–1,920 of 17,313← PreviousNext →