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 |
|---|---|---|---|
| Hear, Localize, and Reason: Spatially Aware Scene Understanding for Audio-Visual LLMs Releases a spatial audio-visual benchmark and model architecture designed to identify, localize, and reason about sound sources in scenes. Ranked #314New testPreprint5/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Introduces benchmarks and models for spatially aware multimodal audio-visual scene reasoning. | |
| VALCON: Benchmarking and Enhancing Value Consistency in Large Language Models via Dual-Role Variation The paper introduces VALCON to evaluate and improve value consistency when language models alternate between generator and evaluator roles. Ranked #314Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Examines value alignment discrepancies and inconsistent evaluator behavior in real deployments. | |
| TreeFlex: Agents Navigating Their Own Search Trees The paper proposes a framework where an agent policy navigates a global search tree by alternating between reflection and execution roles. Ranked #314New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 15 Oct, 15:06 | On-topic Presents an agent framework navigating its own search tree across multi-step execution. | |
| Searching for What Matters: Open-Ended Discovery of Multiple Hypotheses with Agentic Data Exploration The paper introduces an open-ended benchmark for evaluating agentic systems on discovering multiple data-supported hypotheses from tabular datasets. Ranked #314New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | On-topic Introduces a benchmark grounded in causal models to evaluate open-ended hypothesis discovery. | |
| FreeSolo: Linear-time Sparse Attention with Zero-Shot Length Extrapolation Introduces a native sparse attention mechanism that combines sliding windows and keypoint selection for linear-time context extrapolation. Ranked #314Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Proposes a linear-time sparse attention mechanism for efficient inference. | |
| Plausible Is Not Enough: Trust and Evidence in AI-Authored Pull Request Review The authors survey software practitioners to analyze differences in trust, evidence requirements, and review practices for AI-authored pull requests. Ranked #315New way to see a problemPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:19 | On-topic Investigates human review trust, verification, and risks of deploying AI code generators. | |
| PiCo: Physics-Informed Concept Probing for Geo-Foundation Models Uses synthetic physics-based satellite observations to probe and explain physical feature sensitivity inside geo-foundation model representations. Ranked #315New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Probes physical concepts in multimodal earth-observation models using synthetic multispectral and visual data. | |
| NepPlural: A Human-Anchored Benchmark for Persona Annotation in Nepali Migration Discourse NepPlural provides an annotated benchmark of Nepali discourse to evaluate how reliably language models capture diverse socio-cultural personas. Ranked #315Surprising resultPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | On-topic Introduces a benchmark evaluating LLM annotation reliability and pluralistic alignment. | |
| HexaBee-4B: Six-Language Translation with Multilingual Answering and Bidirectional Supervision This paper introduces HexaBee-4B, a translation model trained across six languages using joint multilingual answering and bidirectional supervision. Ranked #315Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:18 | On-topic Adapts a compact 4B language model to outperform larger baselines on translation. | |
| OnTrack: Real-Time Monitoring and Intervention in LLM Agent Trajectories via Streaming Structure-Aware Optimal Transport Proposes a streaming optimal transport method to monitor LLM agent trajectories in real time against reference runs. Ranked #315New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Focuses on monitoring and intervening in multi-step LLM agent trajectories in real time. | |
| Clean Data Can Still Carry Backdoors: Support-Persistent Backdoors for Model Reuse Introduces a clean-data backdoor method that persists through standard knowledge distillation by targeting naturally occurring feature subspaces. Ranked #316Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Investigates persistent backdoor risks in models that survive knowledge distillation defenses. | |
| 3DCodeVerse: Exploring 3D Worlds as Code The authors introduce 3DCodeVerse, a dataset and multi-agent framework enabling coding agents to generate, reason over, and verify 3D executable code. Ranked #316New way to see a problemPreprint4/10 suggested | Multimodal models via arxiv · 17 Oct, 06:17 | On-topic Combines 3D spatial representations and visual feedback with code generation. | |
| VASPO: Self-Distillation with Privileged Value Estimation Combines privileged teacher signals with value estimation in policy optimization to provide denser token-level supervision for reasoning tasks. Ranked #316Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Presents self-distillation techniques to train smaller reasoning models. | |
| Terrarium: Growing Personal Assistants via Scalable Synthesis of App Worlds This paper presents a framework to synthesize executable application environments and training tasks from web information for personal language agents. Ranked #316New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | On-topic Constructs an executable training and evaluation suite with outcome-based rubrics for personal assistants. | |
| Recursive Self-Improvement through Multi-Agent Self-Supervision Presents a multi-agent self-supervision method that uses evolutionary workflow optimization and fine-tuning for recursive model self-improvement. Ranked #316New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Proposes an evolutionary multi-agent workflow where models orchestrate, execute, and self-evaluate complex reasoning tasks. | |
| ExecWeave: Cross-Layer Execution Provenance for Observable Agentic Software Presents a system that links agent application traces to operating-system runtime events to establish clear execution provenance. Ranked #317New way to see a problemPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Develops execution provenance to monitor and audit actions taken by autonomous agentic software. | |
| EgoExplorer: A Vision-Motion-Touch Observatory for Human Loco-Manipulation In Situ This paper introduces a 100-hour multimodal dataset capturing egocentric video, body motion, and touch for human loco-manipulation tasks. Ranked #317New testPreprint4/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Presents a multimodal dataset combining egocentric video, body motion, touch, and language descriptions. | |
| EdgeLLM-Nano: Auditing Polar-Guided Key-Cache Compression and Consistency-Anchored Pruning on a 33.6 MB On-Device Language Model This paper audits key-cache compression and weight pruning techniques on a small edge-deployed language model testbed. Ranked #317Surprising resultPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | On-topic Proposes a fidelity evaluation metric for KV-cache compression. | |
| Next-Concept Modeling: Fast, Parallel Coarse-to-Fine Text Generation This paper introduces a coarse-to-fine text generation method using discrete code hierarchies to accelerate parallel sampling over diffusion language models. Ranked #317Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Presents a coarse-to-fine framework achieving faster parallel text generation and efficient inference. | |
| When Should Agents Think? Adaptive Reasoning via Cross-Turn Estimation Proposes a training method to adaptively skip turn-by-turn agent reasoning when previous thoughts remain sufficient for future actions. Ranked #317New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Studies adaptive reasoning over interaction trajectories in LLM agents to optimize multi-step action planning. | |
| Do Emergent Misalignment Models Share a Circuit? This paper shows that fine-tuned language models exhibiting emergent misalignment across different harmful domains share specific attention heads responsible for the behavior. Ranked #318Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:18 | On-topic Investigates causal attention circuits driving emergent misalignment across distinct harmful domains in fine-tuned models. | |
| VITAL: From Visual Motion to Precise Action This paper presents VITAL, an action tokenizer coupling visual motion scaffolding with residual error encoding for autonomous driving decision-making. Ranked #318New way to see a problemPreprint4/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Couples visual motion and actions for multimodal world-action models. | |
| Diagnosing Generalization Boundaries in AI-Based Vulnerability Detection via Calibrated Embedding Distances The paper evaluates embedding distance metrics to detect generalization boundaries and distribution shifts in AI-based code vulnerability models. Ranked #318New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | On-topic Exposes hidden benchmark degradation in code models under out-of-distribution shifts. | |
| Calibrated Typed Decisions for Entity, Relation, and Event Extraction Presents a unified encoder that jointly models type definitions and text for cross-domain entity, relation, and event extraction. Ranked #318Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Presents an efficient 0.6B-parameter unified model for multi-task information extraction. | |
| ReSI: Recursive Safety Improvement toward Resistant and Resilient AI Presents ReSI, a recursive safety framework using cyclical evaluation and updates to improve AI resilience against emergent threats. Ranked #318New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Implements an automated agent research loop that executes successive rounds of evaluation and updates. | |
| FinGREAT: Evaluating LLMs on Financial Graph-guided Retrieval And Typing The paper benchmarks language models on extracting and categorizing double-entry accounting relationships from corporate financial disclosure filings. Ranked #319New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06 | On-topic Introduces FinGREAT, a benchmark evaluating LLMs on accounting relation classification and evidence discovery. | |
| Generative World Simulation with DAO and JING Couples a persistent symbolic world engine with an egocentric generative model for continuous interactive audiovisual simulation. Ranked #319New way to see a problemPreprint4/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Generates interactive first-person audiovisual experiences driven by text instructions and scene state. | |
| PN-Mix: Pair-Norm Guided Mixed-Precision Key Cache Quantization for Efficient Long-Output LLM Inference The paper proposes a pair-norm guided mixed-precision quantization method for key caches during long-output transformer inference. Ranked #319Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Proposes mixed-precision key cache quantization down to 2.26 bits to optimize long-output LLM inference. | |
| Pristine Models The paper introduces an architecture that separates safe and unsafe visual knowledge during pretraining so unsafe branches can be disabled. Ranked #319New way to see a problemPreprint6/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:18 | On-topic Isolates unsafe concepts during model pretraining to prevent generating unsafe visual content. | |
| InterviewPlayground: A Validated Simulation Environment for Evaluating AI Interviewers Presents a simulated participant environment to evaluate multi-turn AI interviewers across market research and social science scenarios. Ranked #319New testPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Evaluates autonomous AI interview agents conducting multi-turn adaptive interactions with users. | |
| AgentSafeEvo: Adversarial Co-Evolution for LLM Agent Safety The paper introduces an adversarial co-evolution reinforcement learning framework to strengthen language model agents against tool-use manipulation attacks. Ranked #320New way to see a problemPreprint6/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 15:06 | On-topic Applies adversarial co-evolution to protect tool-using agents from malicious attacks. | |
| SWE-Nexus: Automated Composition of Software Engineering Tasks from Test Execution The paper introduces an automated pipeline that builds repository-level software engineering benchmarks from test execution footprints. Ranked #320New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06 | On-topic Introduces an automated benchmark dataset for evaluating software models. | |
| H-DDCM: Hadamard Codebooks for Fast Generative Image and Video Compression This paper presents a generative compression framework using implicit randomized Hadamard dictionaries to speed up diffusion-based image and video coding. Ranked #320Preprint0/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Presents an efficient algorithm to reduce computation and memory for generative diffusion models. | |
| DreamTrue: Action-Faithful Robot World Model with Counterfactual Post-Training Introduces a robot world model using image-rendered actions and counterfactual post-training to improve action-following and contact prediction. Ranked #320New way to see a problemPreprint4/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Develops multi-view multimodal robot world models combining visual prediction with physical action inputs. | |
| Forms of LLM-Integrated Applications from LLM-Chats to Autonomous AI Agent System Surveys and taxonomizes recurring architectural patterns across language-model-integrated software applications, ranging from chatbots to autonomous agent systems. Ranked #320New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Systematically categorizes architectural patterns of autonomous AI agents and tool-executing systems. | |
| Memorizing Who, Knowing How: Benchmarking Personal and Procedural Memory for Persistent Assistants The paper introduces a benchmark to evaluate how personal assistants handle interdependent personal and procedural memories over long trajectories. Ranked #321New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06 | On-topic Introduces PraxisMemBench to evaluate personal and procedural agent memory. | |
| Cortex: Measurement-Gated Evolution and Pruning of Coding-Agent Skills and Rules The authors develop a system that systematically tests, evolves, and prunes skill and rule files used by coding agents. Ranked #321New way to see a problemPreprint6/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Improves coding agent deployment reliability by test-gating skill and rule modifications. | |
| WorldGuide: Goal-Directed Video World Model for Procedural Task Execution Presents a closed-loop goal-directed video world model that iteratively plans and realizes atomic steps for procedural tasks. Ranked #321New testPreprint4/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Combines goal instructions and video generation for closed-loop multimodal procedural task execution. | |
| X-AVDD: Cross-Attentive Audio-Visual Dataset Distillation Proposes a memory-efficient cross-attention method for distilling large paired audio-visual datasets into compact synthetic sets. Ranked #321Preprint0/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:28 | On-topic Proposes dataset distillation to significantly reduce training costs and compute budgets for neural networks. | |
| Trajectory-Guided Fault Localization for Agent Skill Evolution Proposes a method linking execution traces to specific skill segments to guide automated revision of software engineering agent skills. Ranked #321New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Addresses automated skill evolution and trajectory localization for LLM agents. | |
| Can LLMs Anticipate Their Cruxes? Explores methods to extract the underlying decisive sub-questions that drive language model decisions across diverse generated outputs. Ranked #322Surprising resultPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Evaluates and extracts reasoning cruxes behind language model decision outputs. | |
| LEGO: A Lifting-Free Approach for Exocentric-to-Egocentric Video Generation Introduces a lifting-free transformer model that directly synthesizes egocentric video from exocentric recordings without explicit 3D reconstruction. Ranked #322New way to see a problemPreprint4/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Proposes a cross-view transformer and diffusion generator framework for video synthesis. | |
| Shortcut Learning in Geospatial Foundation Model Embeddings for Cross-Disaster Damage Assessment Evaluates shortcut learning and out-of-distribution failure in geospatial foundation models on cross-disaster damage assessment tasks. Ranked #322Surprising resultPreprint6/10 suggested | AI safety and reliability in deployment via arxiv · 12 Oct, 16:28 | On-topic Analyzes shortcut learning and out-of-distribution failure modes in deployed disaster assessment models. | |
| Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks Proposes an external memory framework allowing frozen language model agents to learn and iteratively refine natural-language world models compiled into executable code. Ranked #322New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Develops an LLM agent that maintains world models and plans actions via reflection loops. | |
| The Polytopal Neural Network Introduces a neural network architecture enforcing layer-wise polytopal constraints to improve latent representation interpretability without degrading task performance. Ranked #322Preprint0/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:28 | On-topic Enforces polytope-based latent representations to enable effective vector quantization and compressed representations. | |
| When Repeated Measurements Hurt Tabular Foundation Models The paper studies how repeated input measurements can degrade out-of-sample prediction quality in tabular foundation models. Ranked #323Surprising resultPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | On-topic Evaluates failure modes and performance drops of tabular foundation models. | |
| Rubric-CEPR: Self-Evolving Image Editing via Reward-Verified Self-Distillation Proposes a self-evolving image editing method using rubric-guided contrastive verification without external reward models or human annotations. Ranked #323New way to see a problemPreprint4/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Develops instruction-guided vision-language editing using reward-verified self-distillation. | |
| Zebra3D: Efficient Hybrid Foundation Models For Scalable 3D Reconstruction Proposes a hybrid linear and sparse attention framework to accelerate feed-forward 3D foundation models across many views. Ranked #323Preprint0/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:28 | On-topic Proposes layerwise sparse-linear hybrid attention to speed up foundation model inference. | |
| Kernelized Parameter Sensitivity Atlases for Amortized Conditional Interpretability Develops a kernelized parameter sensitivity atlas to amortize input-specific gradient attribution across vision, forecasting, and biology models. Ranked #323New way to see a problemPreprint6/10 suggested | AI safety and reliability in deployment via arxiv · 12 Oct, 16:28 | On-topic Provides a scalable interpretability attribution method to inspect susceptibility to perturbations in deployments. | |
| Narrow and Deep: An Ontology Tower as the Knowledge of an LLM Agent for an Industrial Equipment System Proposes a narrow-and-deep ontology tower to structure equipment knowledge for language model agents operating industrial air-handling systems. Ranked #323New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Evaluates an LLM agent operating industrial energy equipment using domain knowledge ontologies. | |
| MindEval2: Clinically Grounded Patients and Rubrics for Evaluating Language Models in Mental Health Support Presents an evaluation framework combining simulated mental health patients and expert-developed rubrics to assess conversational language models. Ranked #324New testPreprint6/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Evaluates conversational quality and risks in sensitive clinical mental health deployments. | |
| Are You Sure Your LLM Is Sure? Repairing and Auditing Confidence The paper introduces a framework to recalibrate, audit, and safely deploy confidence scores for language model outputs. Ranked #324New way to see a problemPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Presents methods for evaluating, diagnosing, and calibrating LLM confidence scores. | |
| SmolWAM: A Latent World-Action Model for Robotic Manipulation on Edge Devices Presents a compressed latent world-action model enabling faster robotic manipulation inference on resource-constrained edge hardware. Ranked #324Preprint0/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:29 | On-topic Compresses a world-action policy model for efficient edge device inference with reduced FLOPs. | |
| ContourVLA: A Closed-Loop Perception-Action Contour Policy for Generalized Referring Expression Segmentation Recasts referring expression segmentation as a closed-loop policy iteratively refining explicit boundary contours via visuomotor actions. Ranked #324New way to see a problemPreprint4/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Develops a vision-language-action policy for referring expression segmentation. | |
| What Output-Only Review Cannot Verify: Study Contracts for Research Agents Proposes study contracts to register execution obligations and verify research agent outputs against approved protocol parameters. Ranked #324New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Proposes verification contracts to inspect the execution and artifacts of AI research agents. | |
| SceneForesee: Learning Scene Evolution for Spatiotemporal Reasoning The paper introduces a benchmark and framework to evaluate how multimodal models track entities and reason about future scene evolution in videos. Ranked #325New testPreprint4/10 suggested | Multimodal models via arxiv · 15 Oct, 06:17 | On-topic Advances large multimodal models by modeling spatiotemporal scene evolution and visual-entity reasoning over video. | |
| Where Do LLM Ensemble Gains Come From? Decomposing Candidate Value, Diversity, and Aggregation The paper analyzes how language model aggregation gains divide into candidate additions and context effects under different baseline comparison setups. Ranked #325New way to see a problemPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | On-topic Analyzes evaluation methods and reference controls for measuring candidate value in LLM ensembles. | |
| SynAct: Synergistic Active Feedback Learning for LLM Agents This paper proposes an active learning framework that selectively requests expert feedback at high-value decision steps during LLM agent workflows. Ranked #325New way to see a problemPreprint6/10 suggested | AI safety and reliability in deployment via arxiv · 12 Oct, 16:29 | On-topic Addresses failure modes, error correction, and reliable multi-step execution in deployed LLM agents. | |
| On the Design Space of Continuous Diffusion Language Models Investigates training mechanisms in continuous diffusion language models, proposing design modifications for self-conditioning and embedding geometry. Ranked #325Preprint0/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:29 | On-topic Introduces continuous diffusion language model designs aimed at accelerated sampling and cheaper forward passes. | |
| MemTrial: Learning When to Trust Memory in LLM Portfolio Agents Presents a portfolio management agent that uses fractional factorial counterfactual evaluation and Banzhaf values to isolate memory utility from market movements. Ranked #325New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Investigates memory reuse and multi-step decision-making in LLM portfolio management agents. |
1,921–1,980 of 17,313← PreviousNext →