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 |
|---|---|---|---|
| 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 #230New testPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Presents PremiseSQL, a benchmark evaluating premise critique and abstention in SQL generation. | |
| 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 #230Surprising resultPreprint7/10 suggested | AI agents via arxiv · 17 Oct, 06:17 | On-topic Studies emergent coordination and resource negotiation across long-horizon multi-agent systems. | |
| Understanding and Reducing Misalignment in Persuasion Optimization The paper studies how reinforcement learning for persuasion causes models to fabricate evidence and coercive tactics, evaluating strategies to reduce this behavior. Ranked #230Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Analyzes how reinforcement learning creates misalignment and hallucinated evidence in persuasive models. | |
| Sample-Adaptive Zeroth-Order Post-Training This paper proposes a sample-adaptive zeroth-order optimization method that estimates sample importance online to accelerate large language model fine-tuning. Ranked #230New way to see a problemPreprint5/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:29 | On-topic Proposes memory-efficient zeroth-order optimization for fine-tuning LLMs with reduced resource costs. | |
| PreCoG: Preview-Edit for Efficient Video Correction in Generation-Time Introduces a preview-and-edit framework that diagnoses generation failures from low-cost previews and adjusts cached trajectories during video generation. Ranked #230New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Deals with multimodal text-to-video generation, preview feedback, and semantic trajectory editing. | |
| Beyond Context Length: Characterising the Difficulty of Long-Context Reasoning Presents a benchmark that measures long-context reasoning difficulty using evidence scope, dispersion, and reasoning depth. Ranked #231New testPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Introduces LURCH, a benchmark quantifying long-context reasoning difficulty. | |
| 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 #231New way to see a problemPreprint7/10 suggested | AI agents via arxiv · 17 Oct, 06:17 | On-topic Designs step-level timely interventions for multi-step agentic systems and tool interactions. | |
| Unfair and Unsafe: Everyday Bias Triggers Emergent Misalignment The paper finds that fine-tuning language models on polite text containing demographic stereotypes triggers broader safety and truthfulness failures on unrelated tasks. Ranked #231Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Investigates how fine-tuning on biased text triggers safety misalignments on unrelated tasks. | |
| Cost-Aware Mixture-of-Experts Coordination for Model Markets Proposes a market-level mixture-of-experts coordination framework using cost-aware gating and revenue allocation across heterogeneous model services. Ranked #231New way to see a problemPreprint5/10 suggested | Small and efficient models via arxiv · 12 Oct, 16:29 | On-topic Directly tackles cost-aware coordination, latency, and execution costs in model serving. | |
| A4-MR: Abnormality-aware Anatomical Alignment for MR Introduces a 3D vision-language foundation model trained on MRI data with anatomically aligned text and pathology-aware representations. Ranked #231New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Introduces a 3D vision-language foundation model aligning medical images with text reports. | |
| 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 #232Surprising resultPeer-reviewed7/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Evaluates the reliability of probe-based methods for diagnosing reasoning failures. | |
| Same Context, Different Agents: A Behavioral Study of Semantic Layers, Orchestration and Compaction in Data Agents The paper studies how semantic layers, orchestration, and compaction impact the performance and behavior of data agents across benchmarks. Ranked #232Surprising resultPreprint7/10 suggested | AI agents via arxiv · 17 Oct, 06:17 | On-topic Analyzes behavioral orchestration, tool context, and compaction across autonomous data agents. | |
| CAFE: Confining Feature Distillation to Language-Routed Adapters for Compact Multilingual Automatic Speech Recognition The paper proposes confining feature distillation loss to language-routed adapters to balance performance in compact multilingual speech recognition models. Ranked #232Surprising resultPreprint5/10 suggested | Small and efficient models via arxiv · 15 Oct, 15:06 | On-topic Presents an adapter distillation method to build compact, efficient speech models. | |
| Narrative Paradoxes in LLM Inference: Shaping Reasoning Trajectories toward Misalignment Develops a reasoning-based attack that pairs malicious actors with benign contexts to guide language models along harmful multi-step reasoning trajectories. Ranked #232New way to see a problemPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Examines jailbreak vulnerabilities and safety misalignment in multi-step LLM reasoning trajectories. | |
| A Lightweight Velocity Correction Framework for Improving Instruction Following in Few-Step Text-to-Image Generation Introduces a lightweight velocity corrector to improve prompt following in few-step text-to-image generation models without full fine-tuning. Ranked #232New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Addresses prompt instruction following in generative vision-language models. | |
| WEAVE: Weaving Real-World Code Evolution into Dependency-Structured Tasks for Benchmarking and Training Code Agents Introduces a benchmark that structures multi-pull-request code changes into dependency graphs to evaluate long-horizon software engineering agents. Ranked #233New testPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Presents a novel benchmark and dependency-consistent progress metric for agents. | |
| 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 #233New testPreprint7/10 suggested | AI agents via arxiv · 17 Oct, 06:17 | On-topic Investigates skill-policy co-evolution and skill drift in self-improving agents. | |
| The RNN Homunculus: Learning to Gate Long-Term Memory The paper studies gate learning in recurrent networks, showing how adding short-dependency samples helps models learn long-range temporal connections. Ranked #233Surprising resultPreprint5/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Demonstrates efficient gated RNN architectures matching modern models on long-range dependencies. | |
| ProCARE: Real-World Study Automation via Profile-Grounded Evidence Contracts This paper proposes ProCARE, a framework that uses structured contracts to reliably automate healthcare observational studies with language model agents. Ranked #233New way to see a problemPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Addresses reliability, schema verification, and temporal correctness in healthcare data analysis. | |
| StereoEngine: Scaling Stereo Bimanual Robot Data without Stereo Cameras Synthesizes stereo manipulation data from monocular robot demonstrations using estimated depth and video diffusion models. Ranked #233New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Enables scaling stereo data for Vision-Language-Action (VLA) multimodal robotic models. | |
| The World Won't Wait: Evaluating and Teaching Agents Under a Live Clock Introduces a benchmark evaluating language model agents in real-time environments and a two-timescale training method to improve execution speed. Ranked #234Surprising resultPreprint7/10 suggested | AI agents via arxiv · 17 Oct, 06:17 | On-topic Evaluates LLM agents under real-time interactive game environments with live constraints. | |
| AlphaDiana: A Measurement System for Agents in Verifiable Reasoning Tasks The paper introduces a modular measurement system to evaluate how agent frameworks, tools, and memory impact verifiable reasoning tasks. Ranked #234Surprising resultPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Introduces a standardized measurement system to evaluate LLM agent components, behavior, and resource use. | |
| Pollard: Governed Semantic Execution Trees for Framework-Independent AI Agent Control and Replay This paper presents Pollard, a framework-independent agent runtime that uses semantic execution trees for deterministic tracking, governance, and replay. Ranked #234New way to see a problemPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Provides governance, invariant checking, and auditable replay for reliable agent deployment. | |
| Jev-ToM: Evaluating Theory of Mind in Typed Decision Models The paper introduces a benchmark to evaluate how typed decision models handle theory-of-mind reasoning tasks compared to standard generative language models. Ranked #234New testPreprint5/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Shows typed decision models offer lower cost and latency alternatives to generative LLMs. | |
| VersaCamVLA: Camera-Configurable VLA Policies for Robotic Manipulation Proposes a camera-configurable framework for vision-language-action policies that handles variable camera counts and poses without fixed setups. Ranked #234New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Improves vision-language-action (VLA) multimodal architectures for multi-view robot manipulation. | |
| VELD-QA: A Large-Scale Dataset for Visual Evidence Grounding and Question Answering in Long Documents Presents a benchmark with human-annotated visual evidence to evaluate retrieval and question answering across long documents. Ranked #235New testPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Introduces a large-scale benchmark for evaluating visual evidence grounding and QA. | |
| Lexical Reference Frames: Interpreting LM Residual Streams in Native Semantics Introduces lexical reference frames to interpret intermediate hidden states in language models using the model's own remaining layers. Ranked #235New way to see a problemPreprint5/10 suggested | Small and efficient models via arxiv · 17 Oct, 06:17 | On-topic Provides interpretability frames enabling layer skipping and early exits during inference. | |
| Hy-Memory-30B-A3B: Attributed Reinforcement Learning for Long-Term Memory Writing Introduces a compact language model trained with distillation and reinforcement learning to manage long-term memory across multiple external frameworks. Ranked #235New way to see a problemPreprint7/10 suggested | AI agents via arxiv · 17 Oct, 06:17 | On-topic Develops a memory-writing model enabling agents to manage persistent long-term states. | |
| Phantom Merge: When Your Large Language Model Agents Pick One but Tell You About Another This paper identifies and mitigates Phantom Merge, an error where language model agents bind retrieved facts to the wrong entity. Ranked #235New way to see a problemPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Mitigates entity hallucination and incorrect attribution failures in deployed agents. | |
| From What to Which: Decoding Modifier Grounding in Frozen MLLMs Investigates extracting spatial, attribute, and relational modifier grounding directly from frozen multimodal large language model representations using optimal transport. Ranked #235New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Investigates token-level visual grounding and referential understanding in multimodal LLMs. | |
| 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 #236Surprising resultPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Introduces VoteGEO-Bench to measure the impact of generative engine optimization on visibility. | |
| WEAVE: Weaving Real-World Code Evolution into Dependency-Structured Tasks for Benchmarking and Training Code Agents Introduces a benchmark that structures multi-pull-request code changes into dependency graphs to evaluate long-horizon software engineering agents. Ranked #236New testPreprint7/10 suggested | AI agents via arxiv · 17 Oct, 06:17 | On-topic Evaluates long-horizon autonomous code agents on multi-step repository tasks. | |
| What the Model Cannot See: Covert and Provable Text Invisibility in LLMs The paper examines how trigger-wrapped text can be made completely invisible to language model activations and downstream tasks. Ranked #236New way to see a problemPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Analyzes covert supply-chain backdoors that bypass filtering and code review in LLMs. | |
| Agaram 2.5B: The First Tamil-First Language Model Trained from Scratch with Tool Calling. A Playbook for Under-Represented Languages The authors introduce a 2.5-billion-parameter bilingual Tamil-English language model trained from scratch with tool-calling capabilities. Ranked #236Surprising resultPreprint5/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Details training an efficient 2.5B small language model from scratch on low-resource hardware. | |
| SuperNav: An Agentic Navigation System for Any Task in Any Scene Presents an agent harness that pairs frozen multimodal language models with modular navigation skills for general robot navigation tasks. Ranked #236New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Equips pretrained multimodal LLMs with visual-point interfaces for embodied navigation tasks. | |
| 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 #237Surprising resultPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Proposes VALCON benchmark suites to evaluate LLM value consistency across generation and evaluation. | |
| AlphaDiana: A Measurement System for Agents in Verifiable Reasoning Tasks The paper introduces a modular measurement system to evaluate how agent frameworks, tools, and memory impact verifiable reasoning tasks. Ranked #237Surprising resultPreprint7/10 suggested | AI agents via arxiv · 17 Oct, 06:17 | On-topic Analyzes how agent harnesses, tools, and memory impact accuracy across reasoning benchmarks. | |
| Distilling Text LMs into Speech-to-Speech LMs by Matching Internal Text Predictions The paper proposes a knowledge distillation method that transfers text language model capabilities to speech-to-speech models using internal text interfaces. Ranked #237New way to see a problemPreprint5/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Distills capabilities from large text models into speech-to-speech architectures. | |
| Radioactive Feedback: Tracing Models Trained by Label-Only LLM Supervision The paper proposes an evaluation technique that embeds behavioral traces into score-only feedback to trace distilled student models. Ranked #237New way to see a problemPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Traces unauthorized model distillation via behavioral watermarking in reward feedback. | |
| LIVIN: Benchmarking Spatial and Embodied Intelligence in Digital Twins of Lived-In Homes The paper presents LIVIN, a spatial and embodied intelligence benchmark featuring digital twins of 30 lived-in homes with human-reviewed object arrangements. Ranked #237New testPreprint5/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Benchmarks 3D vision, spatial understanding, and embodied multimodal perception. | |
| 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 #238New way to see a problemPreprint7/10 suggested | AI agents via arxiv · 17 Oct, 06:17 | On-topic Addresses context compaction failure and reliability in long-horizon LLM agents. | |
| What Should We Remove? Evaluating the Reliability and Costs of RLVR Data Decontamination The paper audits data decontamination methods on math corpora to quantify false removals and benchmark leakage risks. Ranked #238Surprising resultPreprint7/10 suggested | LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17 | On-topic Directly assesses benchmark contamination and data decontamination policies for models. | |
| PIG: Parallel Inconsistency Guidance for Accelerating Diffusion Language Models The paper introduces a lightweight guidance model to estimate parallel decoding inconsistency and accelerate diffusion language models. Ranked #238Surprising resultPreprint5/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Proposes parallel decoding guidance to accelerate inference in diffusion language models. | |
| Big-Big-SycoBench: A Benchmark to Detect Sycophancy Through Model Responses to Valid vs. Invalid Arguments The paper introduces a benchmark measuring whether language models resist sycophancy when presented with invalid user arguments versus valid critiques. Ranked #238New testPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Evaluates and detects sycophantic behavior affecting model reliability in decisions. | |
| From Video Clips to Creation Trajectory: Sora100K for AI-Native Video Creation The paper introduces Sora100K, a dataset structuring AI-native video creation workflows into multi-turn creation trajectories and editing operations. Ranked #238New testPreprint5/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Introduces a multimodal dataset and VLM-guided pipeline for text-to-video generation and editing workflows. | |
| Geometric Structure Predicts Per-Example Neighborhood Loss Under PCA Truncation The paper analyzes geometric properties of text embeddings to predict how many nearest neighbors an individual example loses under dimensionality reduction. Ranked #239New way to see a problemPreprint5/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Studies dimensionality reduction and compression of text embeddings to optimize storage and inference costs. | |
| OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding This paper introduces a benchmark to evaluate multimodal models on joint audio-visual temporal grounding, spatial grounding, and question answering. Ranked #239New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | On-topic Introduces a benchmark evaluating multimodal video grounding and reasoning across multiple tracks. | |
| REPRO: Reflective Evolutionary Prompt Optimizer for Automated LLM Red-Teaming This paper proposes a black-box evolutionary prompt optimization framework using a small open-weight model to discover adversarial jailbreaks against language models. Ranked #239New way to see a problemPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Automates black-box red-teaming and prompt injection discovery for LLM safety. | |
| MVGraph: Benchmarking LVLMs on Evidence Composition across Multi-View Visual Graph Introduces a benchmark evaluating vision-language models on graph-theoretic reasoning where evidence is split across multiple visual graph views. Ranked #239New testPreprint5/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Evaluates large vision-language models on reasoning across multiple visual graph views. | |
| Scaling Agentic Data for Long-Horizon Terminal Intelligence Introduces a synthetic data generation framework to produce long-horizon task trajectories for terminal agents. Ranked #239New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:28 | On-topic Introduces synthetic data generation specifically to train long-horizon autonomous terminal agents. | |
| Beyond Model Bias: Cross-Entropy Based Rankings in Literary and Academic Texts Remain Stable across Small Language Models This paper studies cross-entropy surprisal metrics, showing that predictability rankings of texts remain consistent across multiple small language models. Ranked #240Surprising resultPreprint5/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Investigates textual predictability and surprisal specifically across small-scale transformer language models. | |
| When Messages Go Wrong: Understanding Communication Faults in LLM-Based Multi-Agent Systems This paper provides an empirical study and taxonomy of communication faults and error propagation across multi-agent language model systems. Ranked #240Surprising resultPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Evaluates systemic reliability, silent failures, and fault propagation in deployed systems. | |
| SoftwareGym: Scaling Computer-Use Agent Training with Diverse Synthesized Software Environments Introduces an automated environment generation pipeline to synthesize diverse runnable desktop GUI software applications for training computer-use agents. Ranked #240New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 12 Oct, 16:29 | On-topic Develops synthesized GUI environments specifically to train and evaluate computer-use agents. | |
| OmniCapBench: A Deep-Structured Evaluation Framework for Fine-Grained Audio-Visual Captioning Introduces an audio-visual caption benchmark using atomic evaluation units to evaluate multimodal large language models reliably. Ranked #240New testPreprint5/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Introduces a benchmark specifically evaluating multimodal audio-visual large language models on captioning. | |
| EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents Presents a simulation benchmark evaluating how embodied agents continually acquire basic skills and recombine them for complex tasks. Ranked #240New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28 | On-topic Presents EmbodiedCLBench, a benchmark designed to evaluate continual learning in embodied agents. | |
| RuleTrek: Benchmarking Agentic Rule Reasoning in Complex Scenarios The paper introduces a multi-domain benchmark to evaluate language model agents on identifying, understanding, and applying domain-specific rules across complex files. Ranked #241New testPreprint6/10 suggested | LLM evaluation and benchmarks via arxiv · 15 Oct, 15:06 | On-topic Introduces RuleTrek, a multi-domain benchmark evaluating LLM agents on rule identification and deduction. | |
| PlanPath: Exploration-Guided Orchestration for Patient-Level Pathology Reasoning The paper proposes an agent framework that plans multi-slide exploration and visual tool use for patient-level pathology visual question answering. Ranked #241New way to see a problemPreprint5/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Addresses multimodal visual question answering over complex medical pathology slides. | |
| Sensitivity or Criterion? Measuring and Moving When Language Models Act This paper introduces a diagnostic framework using signal detection theory to measure when language model agents decide to use tools. Ranked #241New testPreprint7/10 suggested | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | On-topic Measures agent over-action and unwarranted tool execution risks in deployment. | |
| WorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment Evolution WorkRover introduces a framework to train long-horizon cowork agents by recursively evolving contexts, tasks, and verifiers based on execution feedback. Ranked #241New way to see a problemPreprint6/10 suggested | AI agents via arxiv · 15 Oct, 06:17 | On-topic Presents a framework for training long-horizon collaborative agents via interactive environments. | |
| Looped Reasoning Finishes Earlier Than You Think This paper proposes a recurrent architecture that adds rotational degrees of freedom to prevent oscillatory bottlenecks in looped reasoning. Ranked #241New way to see a problemPreprint5/10 suggested | Small and efficient models via arxiv · 15 Oct, 06:17 | On-topic Explores dynamic-depth looped architectures for faster and more efficient reasoning convergence. |
1,501–1,560 of 17,312← PreviousNext →