Every paper checked — nothing hidden

All papers

Every paper checked for brief #4 of 12–19 Oct 2026 — nothing is hidden. Brief papers first.

All17,313 In brief24 More (on-topic)3,333 Not relevant13,754 Not assessed87 Older352 Could not confirm93 Blogs with no paper266 All blogs284

Paper	Topic	Relevance
Structure Tax: How Structured Output affects LLMs Performance Evaluates how schema ordering impacts LLM performance under structured output constraints, showing reasoning-first designs prevent accuracy drops. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Studies structured output formatting and reasoning preservation rather than multi-step agent planning or tools.
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. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Focuses on general agent reasoning across turns rather than software engineering or code generation.
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. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Does not involve multimodal models or vision-language tasks.
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. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on reasoning efficiency rather than safety, jailbreaks, or alignment risks.
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. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Introduces a training method for adaptive reasoning rather than evaluation benchmarks or testing methodologies.
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. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Does not address retrieval-augmented generation or external document grounding.
MAST: Motif-Augmented Diffusion with Search Tree for Spectroscopic Molecular Structure Elucidation Combines motif priors with search-tree diffusion to generate 2D and 3D molecular structures from spectroscopic data. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Applies a method to a chemical benchmark; does not focus on LLM evaluation or benchmark design.
MAST: Motif-Augmented Diffusion with Search Tree for Spectroscopic Molecular Structure Elucidation Combines motif priors with search-tree diffusion to generate 2D and 3D molecular structures from spectroscopic data. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on molecular structure elucidation via diffusion search trees, not multi-step LLM agents or tool use.
MAST: Motif-Augmented Diffusion with Search Tree for Spectroscopic Molecular Structure Elucidation Combines motif priors with search-tree diffusion to generate 2D and 3D molecular structures from spectroscopic data. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on molecular property prediction rather than LLM safety, guardrails, or deployment reliability.
All Verdicts are Not Equal: Rethinking LLM Judge Reliability Audits the reliability of LLM judges across multiple prompt formats, temperatures, and orderings, introducing a trustworthy verdict metric. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Audits general LLM evaluators rather than models writing, debugging, or reviewing code.
All Verdicts are Not Equal: Rethinking LLM Judge Reliability Audits the reliability of LLM judges across multiple prompt formats, temperatures, and orderings, introducing a trustworthy verdict metric. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Focuses on evaluation reliability rather than model compression, quantization, or efficient serving.
All Verdicts are Not Equal: Rethinking LLM Judge Reliability Audits the reliability of LLM judges across multiple prompt formats, temperatures, and orderings, introducing a trustworthy verdict metric. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Audits text-based LLM judges rather than multimodal vision, audio, or document models.
All Verdicts are Not Equal: Rethinking LLM Judge Reliability Audits the reliability of LLM judges across multiple prompt formats, temperatures, and orderings, introducing a trustworthy verdict metric. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Studies evaluation methodology rather than systems where LLMs plan, use tools, or execute tasks.
All Verdicts are Not Equal: Rethinking LLM Judge Reliability Audits the reliability of LLM judges across multiple prompt formats, temperatures, and orderings, introducing a trustworthy verdict metric. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Focuses on LLM evaluator consistency rather than information retrieval or document-grounded generation.
Is Memorization Context-Sensitive? Prefix-Based Extraction Beyond Isolated Prefixes Analyzes training data extraction risks in language models when prompts include contextual conditioning such as retrieved documents. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Investigates memorization extraction rather than multi-step agent planning or tool execution.
Is Memorization Context-Sensitive? Prefix-Based Extraction Beyond Isolated Prefixes Analyzes training data extraction risks in language models when prompts include contextual conditioning such as retrieved documents. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Analyzes training data extraction phenomena rather than proposing evaluation benchmarks.
EvoAlloc: A Self-Evolving Resource Allocation Agent for Efficient Program Evolution Proposes a self-evolving agent that allocates evaluation resources dynamically across candidate programs during iterative code evolution. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Introduces an optimization agent rather than general LLM evaluation benchmarks or assessment techniques.
EvoAlloc: A Self-Evolving Resource Allocation Agent for Efficient Program Evolution Proposes a self-evolving agent that allocates evaluation resources dynamically across candidate programs during iterative code evolution. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on computational efficiency in code search rather than AI safety or reliability.
Recompose and Refine Latent Reasoning Flows for Vision-Language-Action Models Introduces a framework that dynamically retrieves and recomposes prior latent reasoning steps to guide vision-language-action robot policies. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses policy execution in robotics rather than AI deployment safety, alignment, or guardrails.
Recompose and Refine Latent Reasoning Flows for Vision-Language-Action Models Introduces a framework that dynamically retrieves and recomposes prior latent reasoning steps to guide vision-language-action robot policies. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Develops a robot policy architecture rather than LLM evaluation or benchmarking methods.
Recompose and Refine Latent Reasoning Flows for Vision-Language-Action Models Introduces a framework that dynamically retrieves and recomposes prior latent reasoning steps to guide vision-language-action robot policies. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on continuous robotic control via vision-language-action models rather than autonomous LLM agents.
VINCIE-NExT: Unlocking Video Editing from Images via In-Context Modeling Transfers editing capabilities from image pairs to video diffusion models using in-context visual demonstrations across decomposed sub-tasks. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:28	Not relevant Focuses on creative video editing rather than industrial defect detection or inspection.
VINCIE-NExT: Unlocking Video Editing from Images via In-Context Modeling Transfers editing capabilities from image pairs to video diffusion models using in-context visual demonstrations across decomposed sub-tasks. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on visual synthesis rather than AI safety, guardrails, or alignment.
VINCIE-NExT: Unlocking Video Editing from Images via In-Context Modeling Transfers editing capabilities from image pairs to video diffusion models using in-context visual demonstrations across decomposed sub-tasks. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Introduces a video editing diffusion framework rather than an evaluation benchmark.
VINCIE-NExT: Unlocking Video Editing from Images via In-Context Modeling Transfers editing capabilities from image pairs to video diffusion models using in-context visual demonstrations across decomposed sub-tasks. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Deals with generative video editing techniques rather than LLM agent planning or tool use.
A Geometric Approach to Soft Actor-Critic with Zonotopes for Locomotion Learning Applies zonotope geometry to adaptively scale critic pessimism and disagreement handling in soft actor-critic locomotion policies. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Focuses on deep RL optimization techniques rather than LLM evaluation or benchmark design.
A Geometric Approach to Soft Actor-Critic with Zonotopes for Locomotion Learning Applies zonotope geometry to adaptively scale critic pessimism and disagreement handling in soft actor-critic locomotion policies. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Studies locomotion control optimization rather than AI deployment safety or guardrails.
A Geometric Approach to Soft Actor-Critic with Zonotopes for Locomotion Learning Applies zonotope geometry to adaptively scale critic pessimism and disagreement handling in soft actor-critic locomotion policies. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Develops a reinforcement learning actor-critic algorithm rather than LLM-based autonomous agent systems.
A Geometric Approach to Soft Actor-Critic with Zonotopes for Locomotion Learning Applies zonotope geometry to adaptively scale critic pessimism and disagreement handling in soft actor-critic locomotion policies. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Addresses continuous control RL rather than small language models or efficient inference.
Universal Textual Teaching for LLMs Presents a parameter-free distillation framework transferring knowledge between language models via an iteratively updated natural language guide. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Proposes a knowledge transfer framework rather than studying LLM evaluation or benchmark integrity.
Universal Textual Teaching for LLMs Presents a parameter-free distillation framework transferring knowledge between language models via an iteratively updated natural language guide. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on cross-model capability transfer rather than AI safety guardrails or alignment risks.
Universal Textual Teaching for LLMs Presents a parameter-free distillation framework transferring knowledge between language models via an iteratively updated natural language guide. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on teaching and knowledge distillation between models rather than autonomous tool-using agents.
Universal Textual Teaching for LLMs Presents a parameter-free distillation framework transferring knowledge between language models via an iteratively updated natural language guide. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Focuses on universal textual distillation, not LLM code tools.
Using Weisfeiler-Leman Features for Algorithm Selection in Constraint Optimisation Applies Weisfeiler-Leman graph kernel features to improve automated algorithm selection for constraint optimization problems without training GNNs. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Does not address AI safety, alignment, jailbreaks, or deployment reliability risks.
Using Weisfeiler-Leman Features for Algorithm Selection in Constraint Optimisation Applies Weisfeiler-Leman graph kernel features to improve automated algorithm selection for constraint optimization problems without training GNNs. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Applies graph kernel features to constraint optimization rather than LLM agent workflows.
Using Weisfeiler-Leman Features for Algorithm Selection in Constraint Optimisation Applies Weisfeiler-Leman graph kernel features to improve automated algorithm selection for constraint optimization problems without training GNNs. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Addresses constraint programming algorithms rather than efficient neural model compression or inference.
Using Weisfeiler-Leman Features for Algorithm Selection in Constraint Optimisation Applies Weisfeiler-Leman graph kernel features to improve automated algorithm selection for constraint optimization problems without training GNNs. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Focuses on algorithm selection for constraint programming rather than language model evaluation.
Use and Disuse: Intent-Structured Experience Consolidation for Memory and Learning in LLM Agents Introduces a hierarchical memory architecture that consolidates nested agent task intents into reusable knowledge across long horizons. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Focuses on an agent memory architecture rather than LLM benchmarking or evaluation frameworks.
Use and Disuse: Intent-Structured Experience Consolidation for Memory and Learning in LLM Agents Introduces a hierarchical memory architecture that consolidates nested agent task intents into reusable knowledge across long horizons. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses continual learning in agents rather than safety, guardrails, or jailbreak prevention.
LVS: Local View Synthesis from Relative Camera Pose by Reusing Previous Views Synthesizes novel views by reusing nearby rendered RGB-D images and applying pose-guided warping with residual refinement. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:28	Not relevant Addresses novel view synthesis for graphics/AR rather than industrial inspection or manufacturing perception.
LVS: Local View Synthesis from Relative Camera Pose by Reusing Previous Views Synthesizes novel views by reusing nearby rendered RGB-D images and applying pose-guided warping with residual refinement. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on 3D view synthesis rendering rather than LLM agent workflows or planning.
LVS: Local View Synthesis from Relative Camera Pose by Reusing Previous Views Synthesizes novel views by reusing nearby rendered RGB-D images and applying pose-guided warping with residual refinement. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Pertains to computer graphics rendering, not AI safety or deployment robustness.
LVS: Local View Synthesis from Relative Camera Pose by Reusing Previous Views Synthesizes novel views by reusing nearby rendered RGB-D images and applying pose-guided warping with residual refinement. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Deals with 3D Gaussian Splatting and RGB-D rendering rather than multimodal foundation models.
LVS: Local View Synthesis from Relative Camera Pose by Reusing Previous Views Synthesizes novel views by reusing nearby rendered RGB-D images and applying pose-guided warping with residual refinement. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Proposes a graphics rendering framework rather than an evaluation benchmark for models.
An Investigation of Model Coherence: Narrow Finetunes Contradict Themselves Under Resampling Evaluates model coherence using 175 specific questions to measure internal contradictions and identity conflation under answer resampling. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Studies model coherence and finetuning consistency rather than multi-step agent planning or tool execution.
OA-MAP: Evidence-Grounded Multi-Agent Multimodal Framework for Interpretable Knee Osteoarthritis Progression Introduces an evidence-grounded multi-agent framework combining multimodal medical data and clinical literature for knee osteoarthritis progression prediction. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Applies multimodal agents to medical prediction rather than proposing LLM evaluation benchmarks.
OA-MAP: Evidence-Grounded Multi-Agent Multimodal Framework for Interpretable Knee Osteoarthritis Progression Introduces an evidence-grounded multi-agent framework combining multimodal medical data and clinical literature for knee osteoarthritis progression prediction. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Includes human-in-the-loop review but focuses on clinical prediction rather than deployment AI safety.
Q-Shaped Options for Hierarchical Reinforcement Learning Develops an abstraction approach for hierarchical reinforcement learning to balance temporal option boundaries with state-dependent value representation. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on classical reinforcement learning options and abstraction rather than LLM-based agents.
Q-Shaped Options for Hierarchical Reinforcement Learning Develops an abstraction approach for hierarchical reinforcement learning to balance temporal option boundaries with state-dependent value representation. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Develops a reinforcement learning algorithm rather than an LLM evaluation methodology.
Q-Shaped Options for Hierarchical Reinforcement Learning Develops an abstraction approach for hierarchical reinforcement learning to balance temporal option boundaries with state-dependent value representation. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses HRL policy learning without focusing on AI deployment safety, guardrails, or alignment.
Instruction-Conditioned Electromagnetic Spectrum Understanding via Budget-Adaptive Signal Tokenization Introduces a budget-adaptive tokenization framework that balances token allocation with input length for raw RF signal understanding. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Does not address AI safety, guardrails, robustness, or deployment alignment risks.
Instruction-Conditioned Electromagnetic Spectrum Understanding via Budget-Adaptive Signal Tokenization Introduces a budget-adaptive tokenization framework that balances token allocation with input length for raw RF signal understanding. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Addresses multimodal signal tokenization rather than LLM agent planning, tool execution, or workflows.
Instruction-Conditioned Electromagnetic Spectrum Understanding via Budget-Adaptive Signal Tokenization Introduces a budget-adaptive tokenization framework that balances token allocation with input length for raw RF signal understanding. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Focuses on multimodal signal modeling rather than LLM evaluation methodology or benchmark contamination.
Scalable Hierarchical Graph Generation via Soft Community Structure Presents a three-stage hierarchical generation method for large attributed graphs using recursive soft community decomposition. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Studies scalable graph generation algorithms, not language model compression or efficient LLM inference.
Scalable Hierarchical Graph Generation via Soft Community Structure Presents a three-stage hierarchical generation method for large attributed graphs using recursive soft community decomposition. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Focuses on attributed graph generative modeling rather than LLM evaluation or benchmark reliability.
Scalable Hierarchical Graph Generation via Soft Community Structure Presents a three-stage hierarchical generation method for large attributed graphs using recursive soft community decomposition. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on graph generation models rather than LLM agents, tool use, or planning systems.
Scalable Hierarchical Graph Generation via Soft Community Structure Presents a three-stage hierarchical generation method for large attributed graphs using recursive soft community decomposition. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Does not address AI safety, jailbreaking, hallucinations, or deployment guardrails.
Learning to Plan by Looking Back: Hindsight Hierarchies for Training Reasoning Models Proposes a training framework that teaches reasoning models to plan by reverse-engineering solution ideas from hindsight examples. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Does not study model distillation, quantization, pruning, or efficient serving.
Learning to Plan by Looking Back: Hindsight Hierarchies for Training Reasoning Models Proposes a training framework that teaches reasoning models to plan by reverse-engineering solution ideas from hindsight examples. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Introduces a training loop methodology rather than an LLM evaluation benchmark or framework.
Learning to Plan by Looking Back: Hindsight Hierarchies for Training Reasoning Models Proposes a training framework that teaches reasoning models to plan by reverse-engineering solution ideas from hindsight examples. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses mathematical reasoning training rather than safety, guardrails, or deployment risks.

13,021–13,080 of 13,754← Previous Next →