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
Correcting or Continuing? What On-Policy Distillation Learns from Student Errors The study shows that training on-policy distillation exclusively on student errors degrades accuracy rather than improving it. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Analyzes student error distillation rather than establishing evaluation benchmarks.
Correcting or Continuing? What On-Policy Distillation Learns from Student Errors The study shows that training on-policy distillation exclusively on student errors degrades accuracy rather than improving it. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Examines training distillation dynamics, not deployment safety or guardrails.
Correcting or Continuing? What On-Policy Distillation Learns from Student Errors The study shows that training on-policy distillation exclusively on student errors degrades accuracy rather than improving it. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Investigates distillation training dynamics, not autonomous LLM agents.
LatentRemedy: Efficient Remedy For Non-Prefix KV Cache Reuse in Multi-Agent Systems LatentRemedy uses task-conditioned latent tokens to allow safe key-value cache reuse across conflicting tasks in multi-agent systems. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on inference cache efficiency, not deployment safety or jailbreaks.
LatentRemedy: Efficient Remedy For Non-Prefix KV Cache Reuse in Multi-Agent Systems LatentRemedy uses task-conditioned latent tokens to allow safe key-value cache reuse across conflicting tasks in multi-agent systems. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Proposes an architectural inference remedy, not an evaluation benchmark.
LatentRemedy: Efficient Remedy For Non-Prefix KV Cache Reuse in Multi-Agent Systems LatentRemedy uses task-conditioned latent tokens to allow safe key-value cache reuse across conflicting tasks in multi-agent systems. Preprint	LLMs for code via arxiv · 15 Oct, 06:17	Not relevant Not focused on code generation or software engineering tasks.
LatentRemedy: Efficient Remedy For Non-Prefix KV Cache Reuse in Multi-Agent Systems LatentRemedy uses task-conditioned latent tokens to allow safe key-value cache reuse across conflicting tasks in multi-agent systems. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17	Not relevant Focuses on KV cache reuse across agents, not retrieval-augmented generation.
LatentRemedy: Efficient Remedy For Non-Prefix KV Cache Reuse in Multi-Agent Systems LatentRemedy uses task-conditioned latent tokens to allow safe key-value cache reuse across conflicting tasks in multi-agent systems. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Concerns text KV cache reuse, not multimodal models.
SHELDON: Reasoning through Dual Search over Hypothesis and Observation Spaces SHELDON introduces a dual-search reasoning framework that independently explores hypotheses and observation spaces to avoid confirmation bias. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on scientific discovery search, not deployment safety or guardrails.
SHELDON: Reasoning through Dual Search over Hypothesis and Observation Spaces SHELDON introduces a dual-search reasoning framework that independently explores hypotheses and observation spaces to avoid confirmation bias. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Proposes a reasoning architecture rather than a benchmark or evaluation framework.
SHELDON: Reasoning through Dual Search over Hypothesis and Observation Spaces SHELDON introduces a dual-search reasoning framework that independently explores hypotheses and observation spaces to avoid confirmation bias. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Studies test-time search and reasoning, not model compression or efficiency.
ChampionKV: Head-Wise Adaptive Budget Allocation for KV Cache Compression ChampionKV adaptively allocates key-value cache memory budgets across attention heads using observation-window concentration proxies. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Deals with inference compression rather than safety, jailbreaks, or hallucination.
ChampionKV: Head-Wise Adaptive Budget Allocation for KV Cache Compression ChampionKV adaptively allocates key-value cache memory budgets across attention heads using observation-window concentration proxies. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Develops a cache compression method rather than evaluation methodology.
ChampionKV: Head-Wise Adaptive Budget Allocation for KV Cache Compression ChampionKV adaptively allocates key-value cache memory budgets across attention heads using observation-window concentration proxies. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on KV cache compression, not multi-step agent planning.
Can Machine Learning Potentials Be More Accurate Than the Methods They Are Trained On? The authors propose methods to train machine learning potentials that transfer across quantum-chemical methods to exceed single-method training accuracy. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant About quantum chemistry machine learning potentials, not language model agents.
Can Machine Learning Potentials Be More Accurate Than the Methods They Are Trained On? The authors propose methods to train machine learning potentials that transfer across quantum-chemical methods to exceed single-method training accuracy. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Concerns molecular physics models, not small language models or inference.
Can Machine Learning Potentials Be More Accurate Than the Methods They Are Trained On? The authors propose methods to train machine learning potentials that transfer across quantum-chemical methods to exceed single-method training accuracy. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Focuses on chemical simulation potentials rather than LLM benchmarks.
Can Machine Learning Potentials Be More Accurate Than the Methods They Are Trained On? The authors propose methods to train machine learning potentials that transfer across quantum-chemical methods to exceed single-method training accuracy. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Addresses quantum simulation accuracy, not AI deployment safety or guardrails.
Scaling Vision Transformers Without Training: Prune-and-Merge Token Compression The paper introduces a training-free token pruning and merging framework with new robustness metrics for vision transformers. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:17	Not relevant Applies token reduction to standard vision models, not industrial inspection.
Scaling Vision Transformers Without Training: Prune-and-Merge Token Compression The paper introduces a training-free token pruning and merging framework with new robustness metrics for vision transformers. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Introduces metrics for token pruning robustness rather than LLM benchmarks.
Scaling Vision Transformers Without Training: Prune-and-Merge Token Compression The paper introduces a training-free token pruning and merging framework with new robustness metrics for vision transformers. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on token reduction efficiency rather than safety, guardrails, or jailbreaks.
Scaling Vision Transformers Without Training: Prune-and-Merge Token Compression The paper introduces a training-free token pruning and merging framework with new robustness metrics for vision transformers. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on token compression in vision transformers, not LLM agents.
High-fidelity Three-Dimensional Flow Field Reconstruction Driven by Turbulent Fluid Dynamics The paper proposes a neural operator and state-space framework for high-fidelity 3D turbulent flow reconstruction from coarse data. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Concerns physics flow reconstruction rather than AI deployment safety risks.
High-fidelity Three-Dimensional Flow Field Reconstruction Driven by Turbulent Fluid Dynamics The paper proposes a neural operator and state-space framework for high-fidelity 3D turbulent flow reconstruction from coarse data. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Focuses on physics simulation rather than LLM benchmarking.
High-fidelity Three-Dimensional Flow Field Reconstruction Driven by Turbulent Fluid Dynamics The paper proposes a neural operator and state-space framework for high-fidelity 3D turbulent flow reconstruction from coarse data. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Does not investigate language model efficiency, distillation, or quantization.
High-fidelity Three-Dimensional Flow Field Reconstruction Driven by Turbulent Fluid Dynamics The paper proposes a neural operator and state-space framework for high-fidelity 3D turbulent flow reconstruction from coarse data. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Addresses fluid dynamics turbulence modeling, not language model agents.
Beyond Black Boxes: Visual Grounding for Text Embeddings The paper investigates transforming text embeddings into visual sketches to improve model explainability and evaluate their utility in text retrieval. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Investigates visual embeddings for retrieval rather than AI deployment safety.
Beyond Black Boxes: Visual Grounding for Text Embeddings The paper investigates transforming text embeddings into visual sketches to improve model explainability and evaluate their utility in text retrieval. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Does not establish benchmarks or evaluation methodologies for LLMs.
Beyond Black Boxes: Visual Grounding for Text Embeddings The paper investigates transforming text embeddings into visual sketches to improve model explainability and evaluate their utility in text retrieval. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on embedding representation via sketches, not AI agents.
inLoRA-MINT: Internal Representations for Auditing Training Data in Domain-adapted LLMs This paper presents an intermediate-representation method to audit whether specific private data was used in LoRA fine-tuning of language models. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on membership inference auditing, not interactive LLM agents.
inLoRA-MINT: Internal Representations for Auditing Training Data in Domain-adapted LLMs This paper presents an intermediate-representation method to audit whether specific private data was used in LoRA fine-tuning of language models. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Evaluates text models, not multimodal vision-language architectures.
inLoRA-MINT: Internal Representations for Auditing Training Data in Domain-adapted LLMs This paper presents an intermediate-representation method to audit whether specific private data was used in LoRA fine-tuning of language models. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Presents a privacy audit technique rather than a general LLM evaluation benchmark.
inLoRA-MINT: Internal Representations for Auditing Training Data in Domain-adapted LLMs This paper presents an intermediate-representation method to audit whether specific private data was used in LoRA fine-tuning of language models. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Audits LoRA fine-tuning rather than addressing efficient inference or quantization.
inLoRA-MINT: Internal Representations for Auditing Training Data in Domain-adapted LLMs This paper presents an intermediate-representation method to audit whether specific private data was used in LoRA fine-tuning of language models. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17	Not relevant Does not concern retrieval-augmented generation or external databases.
inLoRA-MINT: Internal Representations for Auditing Training Data in Domain-adapted LLMs This paper presents an intermediate-representation method to audit whether specific private data was used in LoRA fine-tuning of language models. Preprint	LLMs for code via arxiv · 15 Oct, 06:17	Not relevant Does not address code generation, program repair, or coding assistants.
From Policy Evaluation to Off-Policy Control: Provable Emergence of In-Context Q-Learning The authors prove and demonstrate empirically that a two-layer attention Transformer can learn and implement in-context Q-learning algorithms. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Focuses on reinforcement learning theory rather than efficient inference or distillation.
From Policy Evaluation to Off-Policy Control: Provable Emergence of In-Context Q-Learning The authors prove and demonstrate empirically that a two-layer attention Transformer can learn and implement in-context Q-learning algorithms. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Explores theoretical emergence of Q-learning, not safety or guardrails.
From Policy Evaluation to Off-Policy Control: Provable Emergence of In-Context Q-Learning The authors prove and demonstrate empirically that a two-layer attention Transformer can learn and implement in-context Q-learning algorithms. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Theoretical study of in-context Q-learning emergence in small Transformers, not LLM agents.
From Policy Evaluation to Off-Policy Control: Provable Emergence of In-Context Q-Learning The authors prove and demonstrate empirically that a two-layer attention Transformer can learn and implement in-context Q-learning algorithms. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Does not evaluate language models on practical benchmarks.
Uncertainty Evaluation: Bridging Machine Learning and Measurement Science The paper examines conceptual parallels between measurement science and machine learning uncertainty evaluation to propose a more unified assessment framework. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Discusses measurement theory for ML uncertainty, not autonomous LLM agents.
Uncertainty Evaluation: Bridging Machine Learning and Measurement Science The paper examines conceptual parallels between measurement science and machine learning uncertainty evaluation to propose a more unified assessment framework. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Does not concern model compression, distillation, or efficient inference.
WorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment Evolution WorkRover introduces a framework to train long-horizon cowork agents by recursively evolving contexts, tasks, and verifiers based on execution feedback. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Focuses on environment construction for agent training rather than LLM benchmarking.
WorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment Evolution WorkRover introduces a framework to train long-horizon cowork agents by recursively evolving contexts, tasks, and verifiers based on execution feedback. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Addresses training interactive agents, not deployment safety or security risks.
WorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment Evolution WorkRover introduces a framework to train long-horizon cowork agents by recursively evolving contexts, tasks, and verifiers based on execution feedback. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Does not address small models, pruning, or serving efficiency.
AReaL-RSI: Adaptive Scaling and Coordination of Recursive Self-Improvement This paper introduces an agent coordination framework for scaling recursive self-improvement research teams without updating model weights. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on recursive self-improvement coordination rather than safety or guardrails.
AReaL-RSI: Adaptive Scaling and Coordination of Recursive Self-Improvement This paper introduces an agent coordination framework for scaling recursive self-improvement research teams without updating model weights. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Focuses on agent coordination methods rather than introducing evaluation benchmarks.
ResearchFlow-Eval: Tracing the Execution Gap in End-to-End Multi-Claim Autoresearch This paper introduces an evaluation framework grounded in published papers to assess continuous multi-stage execution in automated research agents. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Does not address model compression, pruning, or efficient inference.
ResearchFlow-Eval: Tracing the Execution Gap in End-to-End Multi-Claim Autoresearch This paper introduces an evaluation framework grounded in published papers to assess continuous multi-stage execution in automated research agents. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on workflow benchmarking rather than safety, guardrails, or alignment.
One-Sided Verification Under Limited Expert Review This paper proposes decoupling error proposing from error validation to filter invalid candidates under budget constraints for expert human review. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on filtering candidate scientific solutions, not autonomous multi-step agents.
LifeSearch: Benchmarking Deep Research Agents for Real-World Life and Productivity This paper introduces a benchmark with expert-written rubrics to evaluate deep research agents on real-world life and productivity tasks. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Evaluates agent task capability rather than safety, jailbreaks, or guardrails.
APT: Feedback-Driven Agentic Post-Training for Text-to-Image Generation This paper introduces an agentic post-training framework that coordinates multiple agents to evaluate and adaptively train text-to-image models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on capability improvement rather than AI safety or guardrails.
APT: Feedback-Driven Agentic Post-Training for Text-to-Image Generation This paper introduces an agentic post-training framework that coordinates multiple agents to evaluate and adaptively train text-to-image models. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:17	Not relevant Focuses on text-to-image generation, not industrial inspection or defect detection.
Chain of Thought Monitoring Can Fail Even When CoT Is Necessary This paper shows chain-of-thought monitoring can fail even when reasoning is necessary, proposing a multi-level hierarchy to measure monitorability. Preprint	LLMs for code via arxiv · 15 Oct, 06:17	Not relevant Does not address software engineering, coding assistants, or program repair.
Chain of Thought Monitoring Can Fail Even When CoT Is Necessary This paper shows chain-of-thought monitoring can fail even when reasoning is necessary, proposing a multi-level hierarchy to measure monitorability. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17	Not relevant Does not investigate retrieval-augmented generation or external document grounding.
Chain of Thought Monitoring Can Fail Even When CoT Is Necessary This paper shows chain-of-thought monitoring can fail even when reasoning is necessary, proposing a multi-level hierarchy to measure monitorability. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Does not involve multimodal models or vision-language tasks.
Chain of Thought Monitoring Can Fail Even When CoT Is Necessary This paper shows chain-of-thought monitoring can fail even when reasoning is necessary, proposing a multi-level hierarchy to measure monitorability. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Studies internal reasoning faithfulness rather than agentic tool use and execution.
Chain of Thought Monitoring Can Fail Even When CoT Is Necessary This paper shows chain-of-thought monitoring can fail even when reasoning is necessary, proposing a multi-level hierarchy to measure monitorability. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Not focused on model size reduction, quantization, or serving efficiency.
A CaLLM Judge is a Better Judge: Mitigating Bias in LLM-as-a-judge This paper presents a post-hoc multicalibration framework to correct systematic biases such as position and model family in LLM evaluations. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Does not address model compression, pruning, or efficient inference.
A CaLLM Judge is a Better Judge: Mitigating Bias in LLM-as-a-judge This paper presents a post-hoc multicalibration framework to correct systematic biases such as position and model family in LLM evaluations. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Addresses evaluation bias mitigation, not multi-step planning or tool use.
When Missingness Follows the Graph: Beyond Feature Propagation for Missing Features This paper formalizes graph-structured missing features where missingness spreads across edges, testing existing graph propagation methods under this setting. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Concerns tabular and graph feature missingness, not deployment safety or guardrails.

13,801–13,860 of 17,313← Previous Next →