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
From Research Goals to Scientific Closure: Evaluating Agents Through Evidence-Supported Physics Research Introduces a benchmark evaluating whether language model agents can independently direct and validate goal-specified physics research investigations. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on agent research competence benchmarking, not deployment guardrails.
CF-AVSS: A Coarse-to-Fine Audio-Visual Speech Separation Framework with Target-Speaker Refinement and Post-Fusion for Real-World AVSE Challenge Introduces a two-stage coarse-to-fine speech separation architecture leveraging visual streams to isolate target speaker audio in noisy environments. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Audio-visual signal separation method, not an AI agent architecture.
CF-AVSS: A Coarse-to-Fine Audio-Visual Speech Separation Framework with Target-Speaker Refinement and Post-Fusion for Real-World AVSE Challenge Introduces a two-stage coarse-to-fine speech separation architecture leveraging visual streams to isolate target speaker audio in noisy environments. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Signal processing speech enhancement, not AI safety or robustness.
CF-AVSS: A Coarse-to-Fine Audio-Visual Speech Separation Framework with Target-Speaker Refinement and Post-Fusion for Real-World AVSE Challenge Introduces a two-stage coarse-to-fine speech separation architecture leveraging visual streams to isolate target speaker audio in noisy environments. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Speech processing model, not an LLM evaluation benchmark.
DINO-FIL: Multi-Layer Transformer Fusion for Generative Inpainting Forgery Localization Combines multi-layer vision transformer representations with a fusion decoder to localize pixel-level generative image inpainting manipulations. Peer-reviewed	Computer vision for industrial applications via arxiv · 17 Oct, 06:17	Not relevant Detects generative image inpainting edits, not industrial defect detection.
DINO-FIL: Multi-Layer Transformer Fusion for Generative Inpainting Forgery Localization Combines multi-layer vision transformer representations with a fusion decoder to localize pixel-level generative image inpainting manipulations. Peer-reviewed	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on image editing forensics, not LLM deployment safety.
DINO-FIL: Multi-Layer Transformer Fusion for Generative Inpainting Forgery Localization Combines multi-layer vision transformer representations with a fusion decoder to localize pixel-level generative image inpainting manipulations. Peer-reviewed	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Image forensic detection model, not an LLM benchmark.
DINO-FIL: Multi-Layer Transformer Fusion for Generative Inpainting Forgery Localization Combines multi-layer vision transformer representations with a fusion decoder to localize pixel-level generative image inpainting manipulations. Peer-reviewed	AI agents via arxiv · 17 Oct, 06:17	Not relevant Image forgery localization model, not an AI agent system.
DINO-FIL: Multi-Layer Transformer Fusion for Generative Inpainting Forgery Localization Combines multi-layer vision transformer representations with a fusion decoder to localize pixel-level generative image inpainting manipulations. Peer-reviewed	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Pure computer vision forensic model, not a multimodal language system.
In the Guise of Correction: How Framing and Cultural Adaptation Erode LLM Safety on Real-World Criminal Cases Presents a benchmark based on real criminal cases showing how correction framings and cultural adaptations bypass safety guardrails. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Focuses on criminal prompt compliance, not coding assistants or software repair.
In the Guise of Correction: How Framing and Cultural Adaptation Erode LLM Safety on Real-World Criminal Cases Presents a benchmark based on real criminal cases showing how correction framings and cultural adaptations bypass safety guardrails. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Investigates safety benchmarks, not model efficiency or compression.
In the Guise of Correction: How Framing and Cultural Adaptation Erode LLM Safety on Real-World Criminal Cases Presents a benchmark based on real criminal cases showing how correction framings and cultural adaptations bypass safety guardrails. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Studies jailbreaking and safety erosion, not autonomous multi-step agents.
In the Guise of Correction: How Framing and Cultural Adaptation Erode LLM Safety on Real-World Criminal Cases Presents a benchmark based on real criminal cases showing how correction framings and cultural adaptations bypass safety guardrails. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Focuses on text safety prompts, not multimodal image/video understanding.
In the Guise of Correction: How Framing and Cultural Adaptation Erode LLM Safety on Real-World Criminal Cases Presents a benchmark based on real criminal cases showing how correction framings and cultural adaptations bypass safety guardrails. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Evaluates safety vulnerabilities, not retrieval-augmented generation systems.
RSIFlow: Adaptive Self-Improvement Workflows with an Evolving Meta-Agent Develops a recursive self-improvement framework where an evolving meta-agent adapts agent workflows based on execution trajectories and feedback. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Focuses on agent self-improvement architecture rather than LLM benchmarking.
RSIFlow: Adaptive Self-Improvement Workflows with an Evolving Meta-Agent Develops a recursive self-improvement framework where an evolving meta-agent adapts agent workflows based on execution trajectories and feedback. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Addresses agent orchestration and self-improvement, not safety or guardrails.
Cross-Embodiment Causal Transfer from Human Videos to Robot Manipulation Introduces a causal-beam transfer framework that extracts relevant subtask representations from human videos to guide robotic manipulation. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Robotics policy transfer method, not an LLM benchmark.
Cross-Embodiment Causal Transfer from Human Videos to Robot Manipulation Introduces a causal-beam transfer framework that extracts relevant subtask representations from human videos to guide robotic manipulation. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Addresses cross-embodiment robotics control, not AI safety or guardrails.
Cross-Embodiment Causal Transfer from Human Videos to Robot Manipulation Introduces a causal-beam transfer framework that extracts relevant subtask representations from human videos to guide robotic manipulation. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:17	Not relevant Robotic manipulation imitation learning, not industrial inspection or defect detection.
Cross-Embodiment Causal Transfer from Human Videos to Robot Manipulation Introduces a causal-beam transfer framework that extracts relevant subtask representations from human videos to guide robotic manipulation. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Robotics video-to-action transfer method, not LLM agents.
Co2In: a Bi-level Memory Incremental Learning Framework with Knowledge Encoding, Consolidation, and Integration Proposes a bi-level working and long-term memory framework to balance knowledge retention and acquisition in incremental learning. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Addresses catastrophic forgetting in continual learning, not deployment safety.
Co2In: a Bi-level Memory Incremental Learning Framework with Knowledge Encoding, Consolidation, and Integration Proposes a bi-level working and long-term memory framework to balance knowledge retention and acquisition in incremental learning. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Continual learning neural architecture, not LLM agent orchestration.
Co2In: a Bi-level Memory Incremental Learning Framework with Knowledge Encoding, Consolidation, and Integration Proposes a bi-level working and long-term memory framework to balance knowledge retention and acquisition in incremental learning. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes an incremental learning framework, not an evaluation benchmark.
CIM-Embed: Learning Instruction-Conditioned Multimodal Embeddings for Clustering Learns instruction-conditioned multimodal representations to dynamically cluster objects based on user-specified partition criteria without fixed cluster counts. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Instruction-conditioned embeddings, not model compression or efficient serving.
CIM-Embed: Learning Instruction-Conditioned Multimodal Embeddings for Clustering Learns instruction-conditioned multimodal representations to dynamically cluster objects based on user-specified partition criteria without fixed cluster counts. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Embedding representation technique, not an LLM agent architecture.
CIM-Embed: Learning Instruction-Conditioned Multimodal Embeddings for Clustering Learns instruction-conditioned multimodal representations to dynamically cluster objects based on user-specified partition criteria without fixed cluster counts. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Presents an embedding clustering benchmark, not an LLM evaluation framework.
CIM-Embed: Learning Instruction-Conditioned Multimodal Embeddings for Clustering Learns instruction-conditioned multimodal representations to dynamically cluster objects based on user-specified partition criteria without fixed cluster counts. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Deals with multimodal clustering embeddings, not deployment safety.
RoleQ: Role-Aligned Scales for Converting Language Models to Ternary Proposes a structural scaling method based on transformer unit roles to convert language models to ternary weights effectively. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on model compression efficiency, not safety or guardrails.
RoleQ: Role-Aligned Scales for Converting Language Models to Ternary Proposes a structural scaling method based on transformer unit roles to convert language models to ternary weights effectively. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Quantization framework for LLMs, not code generation or program repair.
RoleQ: Role-Aligned Scales for Converting Language Models to Ternary Proposes a structural scaling method based on transformer unit roles to convert language models to ternary weights effectively. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Focuses on weight quantization, not retrieval-augmented generation.
RoleQ: Role-Aligned Scales for Converting Language Models to Ternary Proposes a structural scaling method based on transformer unit roles to convert language models to ternary weights effectively. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Focuses on language model weight quantization, not multimodal models.
RoleQ: Role-Aligned Scales for Converting Language Models to Ternary Proposes a structural scaling method based on transformer unit roles to convert language models to ternary weights effectively. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Quantization technique, not about AI agent workflows or planning.
RoleQ: Role-Aligned Scales for Converting Language Models to Ternary Proposes a structural scaling method based on transformer unit roles to convert language models to ternary weights effectively. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Introduces a quantization algorithm rather than an evaluation benchmark.
Provably Efficient Risk-Sensitive Offline Reinforcement Learning via Optimized Certainty Equivalents Develops pessimistic value-iteration algorithms with theoretical suboptimality bounds for offline risk-sensitive reinforcement learning using optimized certainty equivalents. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Theoretical statistical bounds, not an LLM evaluation benchmark.
Provably Efficient Risk-Sensitive Offline Reinforcement Learning via Optimized Certainty Equivalents Develops pessimistic value-iteration algorithms with theoretical suboptimality bounds for offline risk-sensitive reinforcement learning using optimized certainty equivalents. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Addresses offline RL convergence bounds, not model efficiency or compression.
Provably Efficient Risk-Sensitive Offline Reinforcement Learning via Optimized Certainty Equivalents Develops pessimistic value-iteration algorithms with theoretical suboptimality bounds for offline risk-sensitive reinforcement learning using optimized certainty equivalents. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Mathematical decision-theoretic risk formulation, not AI deployment safety.
Provably Efficient Risk-Sensitive Offline Reinforcement Learning via Optimized Certainty Equivalents Develops pessimistic value-iteration algorithms with theoretical suboptimality bounds for offline risk-sensitive reinforcement learning using optimized certainty equivalents. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Theoretical offline RL mathematics, not LLM agent systems.
Anvil: Scaling Embodied Spatial Intelligence through Knowledge Recirculation Presents a unified framework with synthetic data generation and benchmarks to train models on embodied spatial reasoning and actions. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:17	Not relevant General embodied spatial reasoning, not industrial defect or quality inspection.
Anvil: Scaling Embodied Spatial Intelligence through Knowledge Recirculation Presents a unified framework with synthetic data generation and benchmarks to train models on embodied spatial reasoning and actions. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on embodied spatial perception and representations, not agent orchestration.
Anvil: Scaling Embodied Spatial Intelligence through Knowledge Recirculation Presents a unified framework with synthetic data generation and benchmarks to train models on embodied spatial reasoning and actions. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Presents an embodied data and modeling framework, not AI safety guardrails.
Anvil: Scaling Embodied Spatial Intelligence through Knowledge Recirculation Presents a unified framework with synthetic data generation and benchmarks to train models on embodied spatial reasoning and actions. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Introduces a dataset/benchmark for spatial robotics, not language model evaluation.
In-Context Learning Before Specialisation: Theory of Learning in a Solvable Topic Model Develops a solvable topic model theory explaining when sequence models adopt in-context learning rather than specialised memorisation during training. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Studies pretraining sample complexity theoretically, not inference efficiency.
In-Context Learning Before Specialisation: Theory of Learning in a Solvable Topic Model Develops a solvable topic model theory explaining when sequence models adopt in-context learning rather than specialised memorisation during training. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Theoretical statistical physics model, not an LLM benchmark.
In-Context Learning Before Specialisation: Theory of Learning in a Solvable Topic Model Develops a solvable topic model theory explaining when sequence models adopt in-context learning rather than specialised memorisation during training. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Theoretical pretraining sequence model analysis, not AI agent workflows.
In-Context Learning Before Specialisation: Theory of Learning in a Solvable Topic Model Develops a solvable topic model theory explaining when sequence models adopt in-context learning rather than specialised memorisation during training. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Theoretical analysis of in-context learning emergence, not deployment safety.
CoWorld Reasoner: Agentic Multimodal Reasoning with the Evolving Video World Model-Driven Harness Equips multimodal vision-language models with video world model operations as tools to support dynamic spatiotemporal visual reasoning. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Presents a reasoning agent framework, not a benchmark or evaluation methodology.
CoWorld Reasoner: Agentic Multimodal Reasoning with the Evolving Video World Model-Driven Harness Equips multimodal vision-language models with video world model operations as tools to support dynamic spatiotemporal visual reasoning. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on multimodal reasoning capability, not safety guardrails or hallucination mitigation.
Learning Rewards in Strategic Games Analyzes strategic identifiability in multi-agent games, showing when inverse reinforcement learning fails to recover true underlying agent rewards. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Theoretical game-theoretic bounds, not an LLM benchmark framework.
Learning Rewards in Strategic Games Analyzes strategic identifiability in multi-agent games, showing when inverse reinforcement learning fails to recover true underlying agent rewards. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Theoretical multi-agent game formulation, not AI deployment safety.
Learning Rewards in Strategic Games Analyzes strategic identifiability in multi-agent games, showing when inverse reinforcement learning fails to recover true underlying agent rewards. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Theoretical game theory and IRL, not practical LLM agents.
Learning Rewards in Strategic Games Analyzes strategic identifiability in multi-agent games, showing when inverse reinforcement learning fails to recover true underlying agent rewards. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Focuses on inverse RL theory, not model compression or efficiency.
CodeHealth Predicts the Token Cost of Agentic Code Review: A Pre-Registered Study of 465 Agent-Authored Pull Requests Examines whether source code quality metrics predict the token costs of multi-agent pull request review pipelines. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Observational empirical study rather than an LLM evaluation benchmark.
CodeHealth Predicts the Token Cost of Agentic Code Review: A Pre-Registered Study of 465 Agent-Authored Pull Requests Examines whether source code quality metrics predict the token costs of multi-agent pull request review pipelines. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies cost predictors in code review, not safety or guardrails.
BridgeAfford: Bridging Query-Target Gap for Generalizable 3D Affordance Grounding Introduces a framework that aligns interaction regions across objects to improve 3D affordance grounding on unseen objects. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Computer vision grounding technique without AI safety or reliability focus.
BridgeAfford: Bridging Query-Target Gap for Generalizable 3D Affordance Grounding Introduces a framework that aligns interaction regions across objects to improve 3D affordance grounding on unseen objects. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:17	Not relevant 3D object affordance grounding, not industrial quality control or inspection.
BridgeAfford: Bridging Query-Target Gap for Generalizable 3D Affordance Grounding Introduces a framework that aligns interaction regions across objects to improve 3D affordance grounding on unseen objects. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Computer vision affordance grounding method, not an agent workflow.
BridgeAfford: Bridging Query-Target Gap for Generalizable 3D Affordance Grounding Introduces a framework that aligns interaction regions across objects to improve 3D affordance grounding on unseen objects. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Vision model methodology rather than an LLM evaluation framework.
Agentic-Priors: Using Large Language Models to Address the Cold Start Problem in Preference Elicitation Uses large language models to construct prior trade-off preferences to address cold-start challenges in choice-learning systems. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Evaluates preference elicitation methods rather than serving as an LLM benchmark.
Agentic-Priors: Using Large Language Models to Address the Cold Start Problem in Preference Elicitation Uses large language models to construct prior trade-off preferences to address cold-start challenges in choice-learning systems. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on Bayesian preference elicitation rather than safety or guardrails.
Following Delegation Traces to Study Human Oversight in Interaction with Multimodal AI Investigates how user trust influences delegation and verification choices during human interaction with multimodal video question-answering systems. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Study of human oversight and trust, not agent planning/tool architectures.

1,981–2,040 of 13,754← Previous Next →