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
Growing and Transferring Pathways for LLM Reinforcement Learning Proposes structural growth pathways to preserve pretrained weights during reinforcement learning and enable transfer across language models. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on RL pathway adaptation, not multi-step tool-using agents.
Growing and Transferring Pathways for LLM Reinforcement Learning Proposes structural growth pathways to preserve pretrained weights during reinforcement learning and enable transfer across language models. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Proposes structural RL pathways, not LLM evaluation protocols.
Drafting Agents, Not Just Tokens: Speculative Recursive Self-Improvement for Agentic Systems Presents speculative recursive self-improvement to draft successor agents before full evaluation and optimize verification budgets. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Optimizes agent search and verification budgets, not model parameter compression.
Drafting Agents, Not Just Tokens: Speculative Recursive Self-Improvement for Agentic Systems Presents speculative recursive self-improvement to draft successor agents before full evaluation and optimize verification budgets. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Introduces an agent optimization framework rather than an evaluation benchmark.
Drafting Agents, Not Just Tokens: Speculative Recursive Self-Improvement for Agentic Systems Presents speculative recursive self-improvement to draft successor agents before full evaluation and optimize verification budgets. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on iterative performance search, not safety guardrails or robustness.
From Caricature to Collective Artifact: How Persona Construction Drives Polarization in Generative Social Simulations Analyzes how persona construction in generative social simulations causes network segregation and collective political polarization. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Studies social simulation dynamics, not standard LLM evaluation benchmarks.
From Caricature to Collective Artifact: How Persona Construction Drives Polarization in Generative Social Simulations Analyzes how persona construction in generative social simulations causes network segregation and collective political polarization. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Examines generative social simulation artifacts, not production AI deployment safety.
pi-Net: Empirical Latent Representations for Differentiable Physics Heads in Data-Scarce Systems Introduces a physics-informed latent variable model combining dimensional analysis and differentiable physics heads for data-scarce multi-physics systems. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on welding multi-physics modeling, not AI safety or guardrails.
pi-Net: Empirical Latent Representations for Differentiable Physics Heads in Data-Scarce Systems Introduces a physics-informed latent variable model combining dimensional analysis and differentiable physics heads for data-scarce multi-physics systems. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Addresses sample efficiency in physical systems, not LLM inference compression.
pi-Net: Empirical Latent Representations for Differentiable Physics Heads in Data-Scarce Systems Introduces a physics-informed latent variable model combining dimensional analysis and differentiable physics heads for data-scarce multi-physics systems. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Proposes physics-informed neural networks, not LLM agent systems.
pi-Net: Empirical Latent Representations for Differentiable Physics Heads in Data-Scarce Systems Introduces a physics-informed latent variable model combining dimensional analysis and differentiable physics heads for data-scarce multi-physics systems. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Develops a physical predictive model, not LLM evaluation benchmarks.
Boosting the Learning for Ranking Patterns Uses analytic hierarchy processes and active learning to rank exploratory data patterns according to individual user preferences efficiently. Peer-reviewed	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Pertains to pattern mining ranking functions, not LLM benchmark evaluation.
Boosting the Learning for Ranking Patterns Uses analytic hierarchy processes and active learning to rank exploratory data patterns according to individual user preferences efficiently. Peer-reviewed	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Covers classical pattern ranking algorithms, not AI deployment safety.
Boosting the Learning for Ranking Patterns Uses analytic hierarchy processes and active learning to rank exploratory data patterns according to individual user preferences efficiently. Peer-reviewed	AI agents via arxiv · 15 Oct, 06:17	Not relevant Addresses pattern mining and decision-making, not LLM-based autonomous agents.
EAU: Environment-Agent-User Co-evolution for Training User-Centric Agents Co-evolves the user simulator, environment tools, and rewards during training to improve interactive multi-turn language model agents. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17	Not relevant Focuses on interactive agent training environments, not RAG pipelines.
EAU: Environment-Agent-User Co-evolution for Training User-Centric Agents Co-evolves the user simulator, environment tools, and rewards during training to improve interactive multi-turn language model agents. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Deals with user-agent text interactions, not multimodal vision-language models.
EAU: Environment-Agent-User Co-evolution for Training User-Centric Agents Co-evolves the user simulator, environment tools, and rewards during training to improve interactive multi-turn language model agents. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Focuses on RL training co-evolution frameworks rather than evaluation benchmarks.
EAU: Environment-Agent-User Co-evolution for Training User-Centric Agents Co-evolves the user simulator, environment tools, and rewards during training to improve interactive multi-turn language model agents. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on training policy convergence, not safety guardrails or red-teaming.
EAU: Environment-Agent-User Co-evolution for Training User-Centric Agents Co-evolves the user simulator, environment tools, and rewards during training to improve interactive multi-turn language model agents. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Proposes agent training workflows, not model compression or efficient serving.
EAU: Environment-Agent-User Co-evolution for Training User-Centric Agents Co-evolves the user simulator, environment tools, and rewards during training to improve interactive multi-turn language model agents. Preprint	LLMs for code via arxiv · 15 Oct, 06:17	Not relevant Addresses general user-interaction agent training, not software engineering models.
Does the Explanation Drift Before the Guarantee? Label-Free Early Warning on Evolving Graphs Monitors model explanation drift on evolving graphs to provide early warnings before predictive uncertainty guarantees fail. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Studies drift detection monitors, not LLM benchmarking or evaluation metrics.
Does the Explanation Drift Before the Guarantee? Label-Free Early Warning on Evolving Graphs Monitors model explanation drift on evolving graphs to provide early warnings before predictive uncertainty guarantees fail. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Studies conformal coverage on graph streams, not autonomous LLM agents.
PatTransBench: an ipc-stratified, multi-dimensional benchmark for patent machine translation evaluation Presents an evaluation benchmark and multi-dimensional protocol for patent machine translation across Chinese, English, Japanese, and Korean. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on translation accuracy metrics, not AI safety or jailbreak prevention.
PatTransBench: an ipc-stratified, multi-dimensional benchmark for patent machine translation evaluation Presents an evaluation benchmark and multi-dimensional protocol for patent machine translation across Chinese, English, Japanese, and Korean. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17	Not relevant Covers patent translation benchmarking, not retrieval-augmented generation systems.
PatTransBench: an ipc-stratified, multi-dimensional benchmark for patent machine translation evaluation Presents an evaluation benchmark and multi-dimensional protocol for patent machine translation across Chinese, English, Japanese, and Korean. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Tests quantized model variants as a case study, not novel model compression.
PatTransBench: an ipc-stratified, multi-dimensional benchmark for patent machine translation evaluation Presents an evaluation benchmark and multi-dimensional protocol for patent machine translation across Chinese, English, Japanese, and Korean. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Focuses strictly on text-based patent translation, not multimodal models.
PatTransBench: an ipc-stratified, multi-dimensional benchmark for patent machine translation evaluation Presents an evaluation benchmark and multi-dimensional protocol for patent machine translation across Chinese, English, Japanese, and Korean. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Evaluates machine translation pipelines, not multi-step tool-using agents.
PatTransBench: an ipc-stratified, multi-dimensional benchmark for patent machine translation evaluation Presents an evaluation benchmark and multi-dimensional protocol for patent machine translation across Chinese, English, Japanese, and Korean. Preprint	LLMs for code via arxiv · 15 Oct, 06:17	Not relevant Focuses on patent language translation rather than software code generation.
Overtaught, Underlearned: Learner-Centered Distillation from Multi-Agent Discussion Distills multi-agent discussions into a single student model by grouping redundant reasoning traces and revisiting unlearned examples. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Studies knowledge distillation methods, not safety guardrails or deployment risks.
Overtaught, Underlearned: Learner-Centered Distillation from Multi-Agent Discussion Distills multi-agent discussions into a single student model by grouping redundant reasoning traces and revisiting unlearned examples. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Addresses student distillation techniques, not LLM evaluation benchmarks or methods.
Heuristics are Not Enough: Calibrated and Corrected Caching for Video Diffusion Acceleration Accelerates video diffusion models by using closed-loop calibrated caching instead of fixed heuristic thresholds across denoising steps. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:17	Not relevant Focuses on video diffusion caching, not industrial visual inspection or defect detection.
Heuristics are Not Enough: Calibrated and Corrected Caching for Video Diffusion Acceleration Accelerates video diffusion models by using closed-loop calibrated caching instead of fixed heuristic thresholds across denoising steps. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Accelerates video diffusion models rather than multi-step tool-using agents.
Heuristics are Not Enough: Calibrated and Corrected Caching for Video Diffusion Acceleration Accelerates video diffusion models by using closed-loop calibrated caching instead of fixed heuristic thresholds across denoising steps. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Proposes a caching method for generation, not LLM benchmarks or evaluation.
Heuristics are Not Enough: Calibrated and Corrected Caching for Video Diffusion Acceleration Accelerates video diffusion models by using closed-loop calibrated caching instead of fixed heuristic thresholds across denoising steps. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Optimizes video diffusion inference speed, not safety, guardrails, or alignment.
Making Language the Low-Rate Code of Vision: Successive Refinement from Words to Pixels Applies rate-distortion theory to structure ordered visual codes so text captions form the initial low-rate representations for images. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Focuses on image coding representations, not LLM evaluation frameworks.
Making Language the Low-Rate Code of Vision: Successive Refinement from Words to Pixels Applies rate-distortion theory to structure ordered visual codes so text captions form the initial low-rate representations for images. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Studies image compression and captioning codes, not autonomous planning agents.
Making Language the Low-Rate Code of Vision: Successive Refinement from Words to Pixels Applies rate-distortion theory to structure ordered visual codes so text captions form the initial low-rate representations for images. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Deals with rate-distortion coding, not AI deployment safety or robustness.
AI Discovery in Combinatorics: From Problems to Solutions and Back to Bigger Problems Introduces a benchmark of 29 open algebraic combinatorics problems to evaluate algorithm discovery by frontier AI models. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Studies frontier AI discovery rather than model compression or efficient serving.
AI Discovery in Combinatorics: From Problems to Solutions and Back to Bigger Problems Introduces a benchmark of 29 open algebraic combinatorics problems to evaluate algorithm discovery by frontier AI models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Discusses scientific discovery meta-problems rather than deployment safety, hallucinations, or guardrails.
G$^2$WAM: World Action Models for Efficient and Geometry Grounded Image-Goal Navigation Uses a learned scene-coordinate localization module to maintain consistent 3D spatial context for image-goal robot navigation. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Addresses embodied visual navigation rather than language model agent systems.
G$^2$WAM: World Action Models for Efficient and Geometry Grounded Image-Goal Navigation Uses a learned scene-coordinate localization module to maintain consistent 3D spatial context for image-goal robot navigation. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on navigation efficiency rather than AI deployment safety or guardrails.
G$^2$WAM: World Action Models for Efficient and Geometry Grounded Image-Goal Navigation Uses a learned scene-coordinate localization module to maintain consistent 3D spatial context for image-goal robot navigation. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:17	Not relevant Focuses on visual navigation rather than manufacturing or industrial defect detection.
G$^2$WAM: World Action Models for Efficient and Geometry Grounded Image-Goal Navigation Uses a learned scene-coordinate localization module to maintain consistent 3D spatial context for image-goal robot navigation. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Presents a navigation architecture rather than general LLM evaluation benchmarks.
VLA-GAUGE: Systematically Evaluating Generalization and Robustness in Vision-Language-Action Models Introduces a diagnostic evaluation framework to systematically test generalization and perturbation robustness in vision-language-action robotic models. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on robotic manipulation models rather than software LLM agent workflows.
VLA-GAUGE: Systematically Evaluating Generalization and Robustness in Vision-Language-Action Models Introduces a diagnostic evaluation framework to systematically test generalization and perturbation robustness in vision-language-action robotic models. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:17	Not relevant Evaluates general robotic manipulation rather than industrial visual inspection or defect detection.
Double-Rollout Reinforcement Learning with Sampling-based Reward Estimation Improves reinforcement learning for reasoning models by sampling continuations from truncated traces to provide more accurate intermediate rewards. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Proposes a training optimization algorithm rather than an evaluation benchmark.
Double-Rollout Reinforcement Learning with Sampling-based Reward Estimation Improves reinforcement learning for reasoning models by sampling continuations from truncated traces to provide more accurate intermediate rewards. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on training reward estimation rather than deployment safety guardrails or alignment.
Double-Rollout Reinforcement Learning with Sampling-based Reward Estimation Improves reinforcement learning for reasoning models by sampling continuations from truncated traces to provide more accurate intermediate rewards. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Addresses RL training for reasoning models rather than multi-step agent tool use.
Towards Modality-Agnostic Seizure Classification: A Unified Deep Learning Framework for Heterogeneous EEG and Intracranial EEG Signals Presents a deep learning framework that integrates heterogeneous surface EEG and intracranial EEG signals for multimodal seizure classification. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Addresses clinical signal processing rather than small efficient LLM serving.
Towards Modality-Agnostic Seizure Classification: A Unified Deep Learning Framework for Heterogeneous EEG and Intracranial EEG Signals Presents a deep learning framework that integrates heterogeneous surface EEG and intracranial EEG signals for multimodal seizure classification. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Evaluates medical biosignals rather than language models or LLM benchmarks.
Towards Modality-Agnostic Seizure Classification: A Unified Deep Learning Framework for Heterogeneous EEG and Intracranial EEG Signals Presents a deep learning framework that integrates heterogeneous surface EEG and intracranial EEG signals for multimodal seizure classification. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on EEG signal classification rather than LLM agents.
Towards Modality-Agnostic Seizure Classification: A Unified Deep Learning Framework for Heterogeneous EEG and Intracranial EEG Signals Presents a deep learning framework that integrates heterogeneous surface EEG and intracranial EEG signals for multimodal seizure classification. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on seizure detection rather than AI deployment safety or guardrails.
MixLID: Token-Level Language Identification in Mixed-Language Text Classifies languages at the token level across 111 language-script combinations in mixed-language and code-switched texts. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Presents a language classification model rather than LLM evaluation methodology.
MixLID: Token-Level Language Identification in Mixed-Language Text Classifies languages at the token level across 111 language-script combinations in mixed-language and code-switched texts. Preprint	LLMs for code via arxiv · 15 Oct, 06:17	Not relevant Applies to multilingual human text rather than software code or coding assistants.
MixLID: Token-Level Language Identification in Mixed-Language Text Classifies languages at the token level across 111 language-script combinations in mixed-language and code-switched texts. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Addresses text-only language identification without multimodal inputs.
MixLID: Token-Level Language Identification in Mixed-Language Text Classifies languages at the token level across 111 language-script combinations in mixed-language and code-switched texts. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on token-level language identification rather than LLM agents.
MixLID: Token-Level Language Identification in Mixed-Language Text Classifies languages at the token level across 111 language-script combinations in mixed-language and code-switched texts. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Focuses on multilingual classification accuracy rather than model compression or inference efficiency.
MixLID: Token-Level Language Identification in Mixed-Language Text Classifies languages at the token level across 111 language-script combinations in mixed-language and code-switched texts. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17	Not relevant Deals with language identification without any retrieval-augmented generation aspects.
MixLID: Token-Level Language Identification in Mixed-Language Text Classifies languages at the token level across 111 language-script combinations in mixed-language and code-switched texts. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Covers language identification rather than AI safety, red-teaming, or guardrails.
When 1+1 Backfires: Rethinking Relevance-Diversity Balancing for Sample-Aware Visual Token Reduction Dynamically balances relevance and visual diversity per sample to compress visual tokens in vision-language models more effectively. Peer-reviewed	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Targets inference efficiency rather than deployment safety, hallucinations, or guardrails.

9,421–9,480 of 13,754← Previous Next →