Every paper checked — nothing hidden
All papers
Every paper checked for brief #4 of 12–19 Oct 2026 — nothing is hidden. Brief papers first.
All17,312In brief24More (on-topic)3,333Not relevant13,753Not assessed87Older351Could not confirm93Blogs with no paper266All blogs284
| Paper | Topic | Relevance | Feedback |
|---|---|---|---|
| Simplifying and Improving On-Policy Distillation for Few-Step Diffusion Proposes Simplified Matching Distillation, a regression-based on-policy method that enables four-step generation for diffusion models. Ranked #87New way to see a problemPreprint6/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:18 | On-topic Introduces an efficient distillation technique to drastically reduce diffusion model sampling steps and cost. | |
| Quality, Not Just Outcome: A Structural Scoring Framework for Selecting Coding-Agent Trajectories for Supervised Fine-Tuning Presents a structural scoring framework to select high-quality coding agent trajectories for supervised fine-tuning based on interaction efficiency. Ranked #87New way to see a problemPreprint7/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Examines trajectory quality metrics to fine-tune autonomous coding agents. | |
| Quality, Not Just Outcome: A Structural Scoring Framework for Selecting Coding-Agent Trajectories for Supervised Fine-Tuning Presents a structural scoring framework to select high-quality coding agent trajectories for supervised fine-tuning based on interaction efficiency. Ranked #87New way to see a problemPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28 | On-topic Proposes a structural trajectory quality scoring system without relying on LLM-as-a-judge. | |
| The Self-Improvement Problem The paper formalizes the theoretical limits of automated gates evaluating self-improvement updates in autonomous agent code. Ranked #88New way to see a problemPreprint8/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Formalizes safety and reliability bounds on autonomous agent self-modification. | |
| State-Aware Generation of Deployable Websites from Visual Prototypes and Requirements The paper introduces a graph framework to maintain consistency across visual elements, workflows, and code when generating websites from prototypes. Ranked #88New way to see a problemPreprint6/10 suggested | LLMs for code via arxiv · 15 Oct, 06:19 | On-topic Presents a method for multimodal models to generate full-stack deployable code from prototypes and requirements. | |
| Simplifying and Improving On-Policy Distillation for Few-Step Diffusion Proposes Simplified Matching Distillation, a regression-based on-policy method that enables four-step generation for diffusion models. Ranked #88New way to see a problemPreprint6/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Applies few-step distillation across multimodal generation tasks including audio-video and 3D synthesis. | |
| Local Citation Recommendation Without Graph Signals The paper introduces a dataset and citation recommendation system that retrieves relevant scientific papers without relying on citation graphs. Ranked #88New testPreprint4/10 suggested | Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:18 | On-topic Introduces a dataset and methods for sentence-level reference retrieval. | |
| EvoCowork: A Benchmark for Real-World Office Tasks This paper introduces a 240-task benchmark to evaluate recursive self-improvement in agents across diverse real-world office workflows. Ranked #88New testPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | On-topic Presents a benchmark and protocol for evaluating harness-level agent self-improvement. | |
| PrivacyGuard: Source-Linked Sanitization and Locally Resolvable PII References for Cloud GUI Agents The paper presents PrivacyGuard, a framework using source-linked sanitization and local references to prevent personal data leakage in cloud GUI agents. Ranked #88New way to see a problemPreprint7/10 suggested | AI agents via arxiv · 15 Oct, 06:17 | On-topic Secures cloud GUI agents taking actions across multi-turn interactions. | |
| Contextual Correction of Parallel Draftsfor Speculative Decoding Introduces a lightweight correction mechanism to restore token dependencies during parallel draft generation in speculative decoding. Ranked #88Surprising resultPreprint6/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:28 | On-topic Develops a contextual correction technique for speculative decoding to accelerate LLM inference throughput. | |
| Orbia: Do Generated Videos Form a Consistent 3D World? Presents a benchmark to evaluate 3D consistency, input preservation, and content persistence in interactive video world models. Ranked #89New testPreprint6/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Evaluates 3D spatial consistency and physical persistence in generative video world models. | |
| Stronger Judges Do Not See More: Deployment-Integrity Blindness in LLM Evaluation Introduces a benchmark measuring how output-only model judges fail to detect tampered runtime environments in agent deployments. Ranked #89Surprising resultPreprint8/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Examines deployment-integrity blindness and security risks where tampered runtime stacks bypass evaluation. | |
| Reviewing Beyond the Change: Repository Size and Existing Code Across Agent and Human Pull Requests The paper studies how often human and agent pull request reviews reference existing repository code outside the submitted change. Ranked #89Surprising resultPreprint6/10 suggested | LLMs for code via arxiv · 15 Oct, 06:19 | On-topic Analyzes how human reviewers interact with code generated by software engineering AI agents. | |
| LongHisDoc: A Comprehensive Benchmark for Chinese Long Historical Document Understanding This paper introduces a benchmark of expert-annotated question-answer pairs to evaluate multimodal comprehension of long historical Chinese documents. Ranked #89New testPreprint4/10 suggested | Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:18 | On-topic Benchmarking retriever performance and end-to-end RAG on long documents. | |
| Hybrid Reasoning Agents The paper proposes an agent architecture that dynamically switches between latent reasoning and text interaction while maintaining external tool access. Ranked #89New way to see a problemPreprint6/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Explores latent reasoning to make agent deliberation more token-efficient than text CoT. | |
| A Distributional Framework for Behavioral Characterization and Risk Management of Large Language Models Presents a Monte Carlo simulation framework to quantify distributional response shifts and model risks across language model lifecycles. Ranked #89New way to see a problemPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29 | On-topic Proposes a Monte Carlo distributional framework to test and characterize stochastic LLM behaviors. | |
| Beyond Binary Verifiable Rewards: Behavioral Self-Rewarding for SWE Agents Introduces behavioral self-rewarding to refine reinforcement learning feedback for software engineering agents by comparing execution trajectories. Ranked #89New way to see a problemPreprint7/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Investigates self-rewarding reinforcement learning strategies specifically to improve software engineering agents. | |
| Can World and Action Modeling Benefit Each Other under Unified Causal Training? Tests whether jointly training action-conditioned world models and robot policy models in a single architecture improves both tasks. Ranked #90Surprising resultPreprint6/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Investigates unified training of vision, video dynamics, and robotic actions. | |
| When Judges Agree and Are Wrong: Consistency-Accuracy Dissociation as a Blind Spot for Label-Free Evaluation Examines failure modes in label-free evaluation where models reach high consistency across perturbations while generating identical incorrect answers. Ranked #90Surprising resultPreprint8/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Highlights hidden failure modes in reliability metrics where models consistently produce identical wrong answers. | |
| Executable but Unfaithful: Mock Object, Stub, and Assertion Generation by LLMs and Coding Agents The paper introduces a benchmark to evaluate how accurately language models and coding agents generate mock objects, stubs, and assertions. Ranked #90New testPreprint6/10 suggested | LLMs for code via arxiv · 15 Oct, 06:19 | On-topic Evaluates LLMs and coding agents on generating software mocks and stubs. | |
| EvoCowork: A Benchmark for Real-World Office Tasks This paper introduces a 240-task benchmark to evaluate recursive self-improvement in agents across diverse real-world office workflows. Ranked #90New testPreprint7/10 suggested | AI agents via arxiv · 15 Oct, 06:17 | On-topic Evaluates recursive self-improvement and procedural performance of agents on office workflows. | |
| FASHION-RQ: Mining Hierarchical Semantic Structure for Automated Evaluation of Fashion Retrieval Introduces FASHION-RQ, an automated framework deriving hierarchical evaluation benchmarks for fashion retrieval directly from catalog structures. Ranked #90New testPreprint4/10 suggested | Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17 | On-topic Focuses on semantic retrieval, ranking, and fine-grained relevance evaluation for search systems. | |
| When Charts Deceive Web Agents: A Paired Benchmark for Misleading Visualization-Grounded Tasks Introduces a benchmark evaluating how misleading visualizations and charts affect the multi-step execution and decisions of web agents. Ranked #90New testPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28 | On-topic Presents MisVisAgentBench, an executable benchmark measuring agent task performance and failure modes. | |
| Surfing the Activation Wave: Analyzing the Information and Geometry of LLM Hidden States for Speculative Decoding Proposes matrix-entropy and kernel alignment criteria to select optimal layer subsets of hidden states for speculative decoding. Ranked #90New way to see a problemPreprint6/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:28 | On-topic Proposes multi-layer selection to optimize speculative decoding for efficient model inference. | |
| Screwtape: Adversarial Transcript Editing for Monitor Red-Teaming The paper introduces an adversarial editing pipeline that alters agent execution transcripts to red-team action-monitoring safety systems. Ranked #91Surprising resultPreprint8/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Addresses safety guardrails and monitoring evasion in deployed AI agents. | |
| Rethinking SZZ Search Strategies: Lessons from Claude Code The paper studies whether flexible search strategies used by coding agents improve the identification of commits that introduce software bugs. Ranked #91Surprising resultPreprint6/10 suggested | LLMs for code via arxiv · 15 Oct, 06:19 | On-topic Analyzes Claude Code's repository search strategies for bug-inducing commit identification. | |
| InfoVis: Unified Slide Image Generation and Editing The paper introduces InfoVis, a 9B multimodal model along with a benchmark and training dataset for slide image generation and editing. Ranked #91New testPreprint6/10 suggested | Multimodal models via arxiv · 15 Oct, 06:17 | On-topic Unifies text rendering, visual grounding, and image generation for slides. | |
| JevGPT: Thinking Fast and Slow in Autonomous Research JevGPT combines fast local evidence assessments with deliberative planning to improve autonomous research agents operating under fixed retrieval budgets. Ranked #91Preprint0/10 suggested | Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17 | On-topic Couples fast evidence retrieval with deliberative integration over external document passages. | |
| DataSense-Bench: The First Step Toward an AI Scientist The paper introduces DataSense-Bench to evaluate whether AI agents can select and rank effective training data subsets for fine-tuning. Ranked #91New testPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29 | On-topic Proposes a new benchmark evaluating AI agents on selecting training data for fine-tuning. | |
| Rethinking Relative Policy Differences Transfer of On-Policy Distillation Formulates relative policy difference transfer as an implicit reward in KL-regularized optimization for flexible on-policy knowledge distillation. Ranked #91Surprising resultPreprint6/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:29 | On-topic Analyzes on-policy distillation to effectively transfer capabilities from teacher models to smaller targets. | |
| When Charts Deceive Web Agents: A Paired Benchmark for Misleading Visualization-Grounded Tasks Introduces a benchmark evaluating how misleading visualizations and charts affect the multi-step execution and decisions of web agents. Ranked #91New testPreprint7/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Directly assesses autonomous Web agents executing multi-step browsing and reasoning workflows. | |
| INFUSE: A Unified Surface Benchmark for Indirect Prompt Injection in Agentic Systems The paper introduces an end-to-end benchmark to evaluate indirect prompt injection risks across long-horizon agent workflows. Ranked #92New testPreprint8/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Benchmarks indirect prompt injection attacks against LLM agents across diverse execution surfaces. | |
| Agentic Evolution of SQL Test Generators The authors use coding agents and execution validation to add new database dialect features into reusable database test generators. Ranked #92New way to see a problemPreprint6/10 suggested | LLMs for code via arxiv · 15 Oct, 06:19 | On-topic Uses coding agents with witness replay to evolve DBMS test generators. | |
| Latent World Action Model for GUI Agents The paper presents a latent world action model that predicts action-induced state transitions rather than full-screen pixel reconstructions for graphical agents. Ranked #92New way to see a problemPreprint6/10 suggested | Multimodal models via arxiv · 15 Oct, 06:17 | On-topic Leverages large multimodal models for visual GUI screen understanding and action modeling. | |
| When Pleasing the User Compromises Care: AI Sycophancy in Caregiver-Mediated Healthcare The paper benchmarks how caregiver-directed language models compromise medical advice when users express clinically inferior care preferences. Ranked #92Surprising resultPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | On-topic Introduces an evidence-grounded benchmark to measure LLM sycophancy when facing harmful user preferences. | |
| PANINI-DR: Controlled Deep Research through Generative Semantic Workspaces Presents a multi-agent framework that decouples research planning, subquestion investigation, and factual verification via a shared semantic workspace. Ranked #92Preprint0/10 suggested | Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17 | On-topic Focuses on deep research workflows involving multi-step search, retrieval, and evidence checking. | |
| Ask Your Advisor: Learning When to Consult Stronger Models in Agentic Systems Develops a reinforcement-learned meta-controller for execution agents to decide when to consult stronger advisor language models cost-effectively. Ranked #92New way to see a problemPreprint7/10 suggested | AI agents via arxiv · 12 Oct, 16:29 | On-topic Studies executor-advisor coordination and dynamic consultation in LLM agent workflows. | |
| One Model to Loop Them All: Elastic Latent Recurrence for Language Models Presents a training method for latent recurrent language models that avoids performance penalties at single loops while scaling with compute. Ranked #92Surprising resultPreprint6/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:28 | On-topic Proposes Elastic Latent Recurrence to dynamically trade inference compute for accuracy without latency penalties. | |
| Velox: Spatial Memory Makes World Models Efficient Introduces a geometry-based attention router that selects visual memory contexts to speed up chunked video world model generation. Ranked #93Surprising resultPreprint6/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Improves long-rollout efficiency in generative video world models using spatial memory routing. | |
| Making every stage of medical device document assessment with large language models deterministic The paper evaluates output non-determinism in commercial language model deployments during medical device regulatory compliance assessments. Ranked #93Surprising resultPreprint8/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Investigates reproducibility and reliability failures of LLMs in high-stakes regulatory compliance deployments. | |
| LLMs program using Reusable and Composable Abstract Representations The paper demonstrates that language models use modular abstract representations to generalize code generation across different programming languages and strategies. Ranked #93New way to see a problemPreprint6/10 suggested | LLMs for code via arxiv · 15 Oct, 15:06 | On-topic Investigates abstract, reusable representations in LLMs during code generation and cross-language transfer. | |
| Joint Subspace Sparse Attention for Efficient Long-Context Decoding The paper introduces a token-level sparse attention method using a learned query-key subspace and an indexed kernel to accelerate long-context decoding. Ranked #93New way to see a problemPreprint6/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Presents a sparse attention method and kernel to improve long-context decoding efficiency. | |
| From Conceptual EneMonIE to an Executable Backend: A Modular Architecture for Energy Information Extractionand Grounded Question Answering The paper describes a modular software backend architecture for extracting energy domain information and answering grounded questions from web sources. Ranked #93Preprint0/10 suggested | Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17 | On-topic Presents an executable backend for grounded question answering from corpus. | |
| AgentStripes: An Evidence-Graded Maturity Model for AI Agents The paper introduces a maturity model that grades artificial intelligence agents across multiple capability levels, evidence stages, and safety gates. Ranked #93New way to see a problemPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | On-topic Proposes AgentStripes, a formal maturity framework and grading methodology for AI agents. | |
| When AI Advisors Agree: Distinguishing Forecast Risk from Forecast Reliability Analyzes archived financial forecasts from multi-model systems to distinguish consensus reliability from baseline underlying asset volatility. Ranked #93Surprising resultPreprint7/10 suggested | AI agents via arxiv · 12 Oct, 16:29 | On-topic Examines multi-LLM advisor agreement, forecast reliability, and risk in real-world deployment. | |
| From Detection to Decision: Calibration and Deferral for Prompt-Injection Detection Evaluates probability calibration and operational allow-block-defer policies for prompt-injection detectors under distribution shifts. Ranked #94Surprising resultPreprint8/10 suggested | AI safety and reliability in deployment via arxiv · 17 Oct, 06:17 | On-topic Evaluates calibration and deferral policies for prompt-injection detection systems. | |
| Rethinking the Role of Tool Graph Memory in Multi-Turn Tool Use The authors analyze tool graph memory in multi-turn agents and find that stored transitions provide little benefit over baselines. Ranked #94Surprising resultPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Analyzes and re-evaluates tool use benchmarks across nine different models. | |
| 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 #94New testPreprint6/10 suggested | LLMs for code via arxiv · 15 Oct, 15:06 | On-topic Constructs repository-level software engineering benchmarks from test execution. | |
| Spectral-Driven Adaptive-Rank Optimizer for Large Language Model Reinforcement Learning with Verifiable Rewards This paper investigates gradient spectral properties during reinforcement learning with verifiable rewards to design an adaptive-rank optimizer for reasoning models. Ranked #94New way to see a problemPreprint6/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Introduces an adaptive low-rank optimizer that exploits low-dimensional gradient structure during training. | |
| How Do Hardware Faults Manifest Differently in LLM Agents? Analyzes how random hardware bit faults propagate and manifest across multi-step LLM agent workflow executions. Ranked #94Surprising resultPreprint7/10 suggested | AI agents via arxiv · 12 Oct, 16:29 | On-topic Evaluates the resilience and failure behavior of tool-using LLM agents under hardware faults. | |
| SPARQL-LLM in the TEXT2SPARQL 2026 Challenge Presents SPARQL-LLM, a multilingual natural language to SPARQL question answering system evaluated in the TEXT2SPARQL 2026 challenge. Ranked #94Peer-reviewed0/10 suggested | Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:29 | On-topic Deals with querying distributed knowledge graphs using LLMs for external knowledge question answering. | |
| DataVista: Diagnosing Multimodal LLMs on Data Video Understanding Introduces a benchmark evaluating multimodal language models on extracting data, tracking narrative flows, and answering questions from animated data videos. Ranked #94New testPreprint6/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Directly assesses multimodal LLMs on animated charts combining visual data, time, and narrative text. | |
| The Code of Theseus: Can Agents Rebuild Software from Recovered Design Documents? The paper extracts software requirements and architecture from code repositories and tests whether language agents can reconstruct projects from them. Ranked #95New testPreprint6/10 suggested | LLMs for code via arxiv · 15 Oct, 15:07 | On-topic Evaluates coding agents rebuilding software from recovered design artefacts and requirements. | |
| GroundGUI: GUI Grounding by High-Quality Data and Distillation-to-Reinforcement Learning Presents a large graphical user interface grounding dataset and a combined distillation and reinforcement learning training method for agents. Ranked #95New testPreprint6/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Develops a multimodal vision-language model and dataset for grounded understanding of screen images. | |
| FlashMask-3: Efficient and Expressive Mask-Aware Distributed Attention The paper presents an expressive and distributed attention kernel that compresses structured attention masks to enable efficient context parallelism on Hopper GPUs. Ranked #95New way to see a problemPreprint6/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:18 | On-topic Designs an efficient mask-aware distributed attention kernel to accelerate long-context training and serving. | |
| Natural Code Transformation Confuses Agents in Vulnerability Analysis The study reports that natural code transformations confuse vulnerability analysis agents, proposing functional changes as a potential defensive strategy. Ranked #95Surprising resultPreprint7/10 suggested | AI agents via arxiv · 15 Oct, 06:17 | On-topic Analyzes how code transformations confuse LLM agents during vulnerability analysis. | |
| Humans vs. Machines: A Comparison With AI Agents in Software Reverse Engineering This paper benchmarks language model agents against skilled humans on binary software reverse engineering tasks using interactive tools. Ranked #95Surprising resultPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | On-topic Compares human analysts against multiple LLM models on standardized software engineering challenges. | |
| Thinking Inertia: LLMs Keep Thinking When Told Not To The paper defines evaluation metrics to show that language models continue generating reasoning traces even when instructed not to think. Ranked #95Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 12 Oct, 16:28 | On-topic Evaluates model unreliability and instruction-following failure when explicitly instructed not to run reasoning traces. | |
| ILM: An AI-Powered Storytelling Educational Tool The paper introduces an educational tool combining Arabic knowledge graphs and passage retrieval to generate narrative-based questions. Ranked #95Preprint0/10 suggested | Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28 | On-topic Combines knowledge graph extraction with passage retrieval to ground question generation and answering. | |
| OS-Bench: Evaluating SWE Agents on Cross-OS Code Migration This paper introduces a benchmark to evaluate software engineering language agents on migrating code between Linux and Windows. Ranked #96New testPreprint6/10 suggested | LLMs for code via arxiv · 15 Oct, 15:07 | On-topic Introduces a benchmark evaluating software engineering agents on cross-OS code migration. |
661–720 of 17,312← PreviousNext →