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 |
|---|---|---|---|
| 4LC: 4D Latent Cache for Efficient Conditioning in Dynamic World Modeling The paper introduces a 4D latent cache to reuse static and dynamic visual representations for efficient conditioning in video generation models. Preprint | Computer vision for industrial applications via arxiv · 15 Oct, 06:17 | Not relevant Targets dynamic video generation, not industrial inspection or defect detection. | |
| 4LC: 4D Latent Cache for Efficient Conditioning in Dynamic World Modeling The paper introduces a 4D latent cache to reuse static and dynamic visual representations for efficient conditioning in video generation models. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Focuses on video generation caching, not language-model planning agents. | |
| 4LC: 4D Latent Cache for Efficient Conditioning in Dynamic World Modeling The paper introduces a 4D latent cache to reuse static and dynamic visual representations for efficient conditioning in video generation models. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Addresses video generation efficiency rather than safety or reliability guardrails. | |
| Executable Prefix Values for Locating Agent Failures The paper evaluates judge labels on agent steps by restarting failed trajectories from intermediate states to identify true failure points. Preprint | Small and efficient models via arxiv · 15 Oct, 06:17 | Not relevant Investigates agent failure analysis rather than model compression or efficiency. | |
| Executable Prefix Values for Locating Agent Failures The paper evaluates judge labels on agent steps by restarting failed trajectories from intermediate states to identify true failure points. Preprint | Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17 | Not relevant Focuses on agent failure trajectory analysis, not retrieval-augmented generation. | |
| Executable Prefix Values for Locating Agent Failures The paper evaluates judge labels on agent steps by restarting failed trajectories from intermediate states to identify true failure points. Preprint | Multimodal models via arxiv · 15 Oct, 06:17 | Not relevant Examines software engineering agent trajectories, not multimodal or vision tasks. | |
| The Tokens Strike Back: Safety-Aware Denoising with Directional Embedding-based Revision Presents SADDER, an inference-time text diffusion safety framework using directional embedding signals to steer unsafe tokens toward benign distributions. Preprint | Small and efficient models via arxiv · 15 Oct, 06:17 | Not relevant Investigates safety steering rather than model compression or inference efficiency. | |
| The Tokens Strike Back: Safety-Aware Denoising with Directional Embedding-based Revision Presents SADDER, an inference-time text diffusion safety framework using directional embedding signals to steer unsafe tokens toward benign distributions. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Introduces a safety steering method, not an evaluation framework. | |
| The Tokens Strike Back: Safety-Aware Denoising with Directional Embedding-based Revision Presents SADDER, an inference-time text diffusion safety framework using directional embedding signals to steer unsafe tokens toward benign distributions. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Focuses on diffusion model sampling, not tool-using or planning agents. | |
| Beyond Bias Scores: Answerability Confounds in Activation Steering with Gradient-Constrained Direction Design (GCD) Shows that activation steering evaluations can conflate bias reduction with model abstention, introducing a gradient-constrained direction design to isolate preferences. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Studies activation steering and bias metrics, not autonomous agent planning. | |
| When Should Enterprise Policy Decisions Be Learned, and When Should They Be Reasoned? Evaluates enterprise policy enforcement across structured JSON versus single-token outputs, measuring the impact of explicit reasoning on decision stability. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Investigates agent decision interfaces rather than establishing a general LLM benchmark. | |
| RoTA-Bench: Evaluating 3D Spatial Reasoning through Interactive Projection Alignment Introduces RoTA-Bench to evaluate vision-language models on 3D spatial reasoning through interactive object rotation to match target 2D silhouettes. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Focuses on 3D spatial reasoning evaluation rather than autonomous agent systems. | |
| RoTA-Bench: Evaluating 3D Spatial Reasoning through Interactive Projection Alignment Introduces RoTA-Bench to evaluate vision-language models on 3D spatial reasoning through interactive object rotation to match target 2D silhouettes. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Evaluates spatial reasoning accuracy, not safety, jailbreaks, or guardrails. | |
| DeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress Meter Introduces DeltaTTG, a method using predicted remaining token count as a progress signal to allocate advantages in multi-turn reinforcement learning. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Aimed at task completion optimization rather than safety or reliability. | |
| DeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress Meter Introduces DeltaTTG, a method using predicted remaining token count as a progress signal to allocate advantages in multi-turn reinforcement learning. Preprint | Small and efficient models via arxiv · 15 Oct, 06:17 | Not relevant Does not focus on model compression, pruning, or efficient inference serving. | |
| DeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress Meter Introduces DeltaTTG, a method using predicted remaining token count as a progress signal to allocate advantages in multi-turn reinforcement learning. Preprint | Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17 | Not relevant Focuses on step-level agent RL, not RAG architecture or retrieval. | |
| DeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress Meter Introduces DeltaTTG, a method using predicted remaining token count as a progress signal to allocate advantages in multi-turn reinforcement learning. Preprint | LLMs for code via arxiv · 15 Oct, 06:17 | Not relevant Applies to multi-turn agents generally, not specifically code assistants. | |
| DeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress Meter Introduces DeltaTTG, a method using predicted remaining token count as a progress signal to allocate advantages in multi-turn reinforcement learning. Preprint | Multimodal models via arxiv · 15 Oct, 06:17 | Not relevant Evaluates text-based agents, not multimodal or vision-language models. | |
| DeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress Meter Introduces DeltaTTG, a method using predicted remaining token count as a progress signal to allocate advantages in multi-turn reinforcement learning. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes an RL training algorithm, not an evaluation methodology or benchmark. | |
| PrT-Flow: Probability-Tractable Flow Matching for Reinforcement Learning Proposes PrT-Flow, an invertible flow matching method that enables exact action probability computation for reinforcement learning in robotics. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Focuses on robotics action density estimation, not AI deployment safety. | |
| PrT-Flow: Probability-Tractable Flow Matching for Reinforcement Learning Proposes PrT-Flow, an invertible flow matching method that enables exact action probability computation for reinforcement learning in robotics. Preprint | Computer vision for industrial applications via arxiv · 15 Oct, 06:17 | Not relevant Addresses general robot motor control benchmarks, not industrial visual inspection. | |
| PrT-Flow: Probability-Tractable Flow Matching for Reinforcement Learning Proposes PrT-Flow, an invertible flow matching method that enables exact action probability computation for reinforcement learning in robotics. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Focuses on continuous robotic reinforcement learning, not LLM-based autonomous agents. | |
| PrT-Flow: Probability-Tractable Flow Matching for Reinforcement Learning Proposes PrT-Flow, an invertible flow matching method that enables exact action probability computation for reinforcement learning in robotics. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Introduces a continuous RL method rather than general model evaluation methods. | |
| Cowork-Gym: Synthesizing Interactive Work Environments at Scale Introduces Cowork-Gym, a framework synthesizing multi-tool collaborative work environments and user simulations to train enterprise agents. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Focuses on agent task capability and benchmark performance, not safety guardrails. | |
| Encounter-Driven City-Scale Human Mobility Generation with Large Language Models Introduces EncounterMob, an LLM-based framework that generates human mobility trajectories while preserving both individual routines and multi-person encounter patterns. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Addresses mobility generation rather than AI safety, guardrails, or alignment. | |
| Encounter-Driven City-Scale Human Mobility Generation with Large Language Models Introduces EncounterMob, an LLM-based framework that generates human mobility trajectories while preserving both individual routines and multi-person encounter patterns. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on human mobility modeling rather than LLM benchmarking standards. | |
| Encounter-Driven City-Scale Human Mobility Generation with Large Language Models Introduces EncounterMob, an LLM-based framework that generates human mobility trajectories while preserving both individual routines and multi-person encounter patterns. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Applies LLMs to mobility generation, not agent planning or tool execution. | |
| Dynamic K-Head: Difficulty-Supervised Budget Selection for Continuous Latent Reasoning Proposes Dynamic K-Head, a question-conditioned budget selector that sets continuous latent reasoning steps based on input difficulty. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a dynamic budget selector rather than a new evaluation framework. | |
| Dynamic K-Head: Difficulty-Supervised Budget Selection for Continuous Latent Reasoning Proposes Dynamic K-Head, a question-conditioned budget selector that sets continuous latent reasoning steps based on input difficulty. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Optimizes latent reasoning budgets, not tool use or autonomous agents. | |
| Dynamic K-Head: Difficulty-Supervised Budget Selection for Continuous Latent Reasoning Proposes Dynamic K-Head, a question-conditioned budget selector that sets continuous latent reasoning steps based on input difficulty. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Focuses on reasoning efficiency, not deployment safety or robustness. | |
| LatentRiskBench: Measuring State-Dependent Risks in Enterprise Agents Introduces LatentRiskBench to evaluate safety risks in enterprise agents where action safety depends on latent workflow context. Preprint | Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17 | Not relevant Focuses on agent workflow state risks rather than retrieval-augmented generation systems. | |
| LatentRiskBench: Measuring State-Dependent Risks in Enterprise Agents Introduces LatentRiskBench to evaluate safety risks in enterprise agents where action safety depends on latent workflow context. Preprint | Multimodal models via arxiv · 15 Oct, 06:17 | Not relevant Focuses entirely on text-based agent actions, not multimodal or vision-language data. | |
| LatentRiskBench: Measuring State-Dependent Risks in Enterprise Agents Introduces LatentRiskBench to evaluate safety risks in enterprise agents where action safety depends on latent workflow context. Preprint | LLMs for code via arxiv · 15 Oct, 06:17 | Not relevant Evaluates enterprise agent workflow actions, not code generation or software engineering. | |
| LatentRiskBench: Measuring State-Dependent Risks in Enterprise Agents Introduces LatentRiskBench to evaluate safety risks in enterprise agents where action safety depends on latent workflow context. Preprint | Small and efficient models via arxiv · 15 Oct, 06:17 | Not relevant Does not explore model compression, efficiency, quantization, or on-device execution. | |
| Stable Token-level Scheduling for Serverless LLM Inference Proposes Token Window Pacing, a serverless scheduling framework that regulates request admissions at the token level to reduce inter-token latency. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Focuses on serverless LLM serving latency, not autonomous planning or tool use. | |
| Stable Token-level Scheduling for Serverless LLM Inference Proposes Token Window Pacing, a serverless scheduling framework that regulates request admissions at the token level to reduce inter-token latency. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Addresses system latency and throughput rather than safety, jailbreaks, or alignment. | |
| Stable Token-level Scheduling for Serverless LLM Inference Proposes Token Window Pacing, a serverless scheduling framework that regulates request admissions at the token level to reduce inter-token latency. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes an inference scheduling algorithm, not an evaluation method or benchmark. | |
| MedEvo: Medical Reasoning Evolution through Knowledge Navigation Proposes MedEvo, an iterative training framework that repairs multimodal medical reasoning trajectories using minimal hints and taxonomy-guided feedback. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Focuses on medical multimodal fine-tuning, not autonomous agents. | |
| MedEvo: Medical Reasoning Evolution through Knowledge Navigation Proposes MedEvo, an iterative training framework that repairs multimodal medical reasoning trajectories using minimal hints and taxonomy-guided feedback. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a training framework rather than an evaluation benchmark. | |
| MedEvo: Medical Reasoning Evolution through Knowledge Navigation Proposes MedEvo, an iterative training framework that repairs multimodal medical reasoning trajectories using minimal hints and taxonomy-guided feedback. Preprint | Small and efficient models via arxiv · 15 Oct, 06:17 | Not relevant Focuses on trajectory repair for medical models, not compression or serving. | |
| MedEvo: Medical Reasoning Evolution through Knowledge Navigation Proposes MedEvo, an iterative training framework that repairs multimodal medical reasoning trajectories using minimal hints and taxonomy-guided feedback. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Focuses on diagnostic task accuracy rather than safety guardrails or reliability. | |
| MedEvo: Medical Reasoning Evolution through Knowledge Navigation Proposes MedEvo, an iterative training framework that repairs multimodal medical reasoning trajectories using minimal hints and taxonomy-guided feedback. Preprint | Computer vision for industrial applications via arxiv · 15 Oct, 06:17 | Not relevant Targets clinical medical imaging, not industrial manufacturing inspection. | |
| HelpMeOut: A Neuropsychology-Grounded Benchmark for Error-Aware Monitoring of Household Task Execution Introduces HelpMeOut, an embodied AI benchmark for diagnosing human execution errors across household activity trajectories motivated by neuropsychology research. Preprint | Computer vision for industrial applications via arxiv · 15 Oct, 06:17 | Not relevant Focuses on household assistance monitoring, not industrial defect inspection. | |
| zPocket: Better Pocket Finding with Frozen Co-Folding Features Introduces the PocketWatch benchmark and zPocket, an approach training pocket-prediction heads on frozen representations from protein co-folding models. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Focuses on computational drug discovery rather than AI deployment safety. | |
| zPocket: Better Pocket Finding with Frozen Co-Folding Features Introduces the PocketWatch benchmark and zPocket, an approach training pocket-prediction heads on frozen representations from protein co-folding models. Preprint | Small and efficient models via arxiv · 15 Oct, 06:17 | Not relevant Focuses on structural biology models, not model compression. | |
| zPocket: Better Pocket Finding with Frozen Co-Folding Features Introduces the PocketWatch benchmark and zPocket, an approach training pocket-prediction heads on frozen representations from protein co-folding models. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Focuses on protein-ligand binding pockets, not LLM agents. | |
| zPocket: Better Pocket Finding with Frozen Co-Folding Features Introduces the PocketWatch benchmark and zPocket, an approach training pocket-prediction heads on frozen representations from protein co-folding models. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates biological pocket prediction, not LLM benchmarks. | |
| LeanPRM: Lean-Grounded Process Supervision for Natural Language Proof Verification Presents LeanPRM, a method using the Lean proof assistant to generate step-level correctness labels for training natural-language mathematical proof verifiers. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Addresses proof verification accuracy rather than deployment safety. | |
| LeanPRM: Lean-Grounded Process Supervision for Natural Language Proof Verification Presents LeanPRM, a method using the Lean proof assistant to generate step-level correctness labels for training natural-language mathematical proof verifiers. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Focuses on mathematical proof step-verification, not multi-step agent actions. | |
| LeanPRM: Lean-Grounded Process Supervision for Natural Language Proof Verification Presents LeanPRM, a method using the Lean proof assistant to generate step-level correctness labels for training natural-language mathematical proof verifiers. Preprint | Small and efficient models via arxiv · 15 Oct, 06:17 | Not relevant Focuses on automated supervision for verifiers, not model compression. | |
| Understanding Gene Expression-to-Image Learning: Spatial Structure and Information Fidelity Studies the trade-off between spatial smoothness and gene-level fidelity when mapping single-cell RNA sequencing profiles to images for neural networks. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Concerns biological data representations rather than AI safety. | |
| Understanding Gene Expression-to-Image Learning: Spatial Structure and Information Fidelity Studies the trade-off between spatial smoothness and gene-level fidelity when mapping single-cell RNA sequencing profiles to images for neural networks. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Focuses on genomics and RNA representations, not LLM agents. | |
| Understanding Gene Expression-to-Image Learning: Spatial Structure and Information Fidelity Studies the trade-off between spatial smoothness and gene-level fidelity when mapping single-cell RNA sequencing profiles to images for neural networks. Preprint | Small and efficient models via arxiv · 15 Oct, 06:17 | Not relevant Focuses on biological representation learning, not model compression. | |
| Understanding Gene Expression-to-Image Learning: Spatial Structure and Information Fidelity Studies the trade-off between spatial smoothness and gene-level fidelity when mapping single-cell RNA sequencing profiles to images for neural networks. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Studies gene-to-image encoding, not LLM benchmark evaluation. | |
| A Node Embedding Framework for Model-Centric Conformance Checking in Process Mining Introduces a graph embedding and Procrustes alignment framework to provide a bounded metric for evaluating discovered process models against ground truth. Preprint | AI safety and reliability in deployment via arxiv · 15 Oct, 06:17 | Not relevant Addresses process mining metrics rather than AI safety or reliability. | |
| A Node Embedding Framework for Model-Centric Conformance Checking in Process Mining Introduces a graph embedding and Procrustes alignment framework to provide a bounded metric for evaluating discovered process models against ground truth. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates business process models, not language models. | |
| A Node Embedding Framework for Model-Centric Conformance Checking in Process Mining Introduces a graph embedding and Procrustes alignment framework to provide a bounded metric for evaluating discovered process models against ground truth. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Presents process mining graph methods, not LLM agents. | |
| From Token Correction to Claim Admission: Training-Free Hallucination Mitigation for Large Vision-Language Models Proposes a training-free framework that mitigates vision-language model hallucinations by admitting claims only when supported relative to a visual-null input. Preprint | AI agents via arxiv · 15 Oct, 06:17 | Not relevant Focuses on vision-language hallucination, not tool-using agents. | |
| From Token Correction to Claim Admission: Training-Free Hallucination Mitigation for Large Vision-Language Models Proposes a training-free framework that mitigates vision-language model hallucinations by admitting claims only when supported relative to a visual-null input. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a hallucination mitigation algorithm rather than a new benchmark. | |
| From Token Correction to Claim Admission: Training-Free Hallucination Mitigation for Large Vision-Language Models Proposes a training-free framework that mitigates vision-language model hallucinations by admitting claims only when supported relative to a visual-null input. Preprint | Computer vision for industrial applications via arxiv · 15 Oct, 06:17 | Not relevant Studies general vision-language hallucination, not industrial defect detection. |
10,021–10,080 of 13,754← PreviousNext →