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
Librarian Agents: Training LLMs to Curate Long-Context Workspaces that Anticipate Future Queries The paper trains language model agents to reorganize unstructured document collections into structured workspaces to speed up future multi-document queries. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on task performance and curation rather than safety or guardrails.
LEASH: Uncertainty-Aware Weak-to-Strong Escalation and Oversight for Computer-Use Agents The paper proposes a training-free framework that uses uncertainty signals to escalate computer-use tasks from weak models to strong models. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Presents an agent routing architecture rather than benchmark evaluation methodologies.
Asking Smarter Questions: Targeted Active Preference Learning for Underspecified Rewards The paper presents a method for robots to generate targeted preference queries to learn underspecified rewards from humans. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:17	Not relevant Applies to robotics preference learning, not industrial manufacturing vision.
Asking Smarter Questions: Targeted Active Preference Learning for Underspecified Rewards The paper presents a method for robots to generate targeted preference queries to learn underspecified rewards from humans. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Addresses reward learning for robotics rather than LLM-based autonomous agents.
Asking Smarter Questions: Targeted Active Preference Learning for Underspecified Rewards The paper presents a method for robots to generate targeted preference queries to learn underspecified rewards from humans. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes an active preference learning algorithm rather than an evaluation benchmark.
Asking Smarter Questions: Targeted Active Preference Learning for Underspecified Rewards The paper presents a method for robots to generate targeted preference queries to learn underspecified rewards from humans. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Does not investigate model compression, distillation, or efficient inference.
Improving Environment-Agnostic Coding Agents: The Role of Scaffolding The authors evaluate how developer scaffolding and general prompt instructions affect the performance of environment-agnostic coding agents across diverse benchmarks. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Focuses on agent scaffolding interventions rather than benchmark design.
Improving Environment-Agnostic Coding Agents: The Role of Scaffolding The authors evaluate how developer scaffolding and general prompt instructions affect the performance of environment-agnostic coding agents across diverse benchmarks. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Investigates agent performance engineering rather than safety or reliability guardrails.
EliteReplay: Preserving Diverse Solutions in RL Post-Training via Quality-Diversity Search The authors introduce EliteReplay, combining reinforcement learning with quality-diversity search replay to retain solution diversity during post-training. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Aims to retain solution diversity rather than mitigate safety failures.
EliteReplay: Preserving Diverse Solutions in RL Post-Training via Quality-Diversity Search The authors introduce EliteReplay, combining reinforcement learning with quality-diversity search replay to retain solution diversity during post-training. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on post-training RL loss design rather than agentic workflows.
EliteReplay: Preserving Diverse Solutions in RL Post-Training via Quality-Diversity Search The authors introduce EliteReplay, combining reinforcement learning with quality-diversity search replay to retain solution diversity during post-training. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Investigates RL optimization techniques rather than evaluation methodologies.
Towards Continual Learning: Mitigating Forgetting With Subspace Tuning The authors propose Subspace Tuning, a fine-tuning method restricting weight updates to bottom singular subspaces to mitigate catastrophic forgetting in language models. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Investigates continual learning fine-tuning methods, not autonomous tool-using agents.
Towards Continual Learning: Mitigating Forgetting With Subspace Tuning The authors propose Subspace Tuning, a fine-tuning method restricting weight updates to bottom singular subspaces to mitigate catastrophic forgetting in language models. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on continual learning stability rather than deployment safety guardrails.
Towards Continual Learning: Mitigating Forgetting With Subspace Tuning The authors propose Subspace Tuning, a fine-tuning method restricting weight updates to bottom singular subspaces to mitigate catastrophic forgetting in language models. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes a tuning method rather than benchmark evaluation frameworks.
Private Retrieval Augmented Generation via Random Projection The authors present a random projection method to protect sensitive datastore embeddings against extraction attacks in retrieval-augmented generation systems. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on RAG datastore privacy, not agentic planning.
Private Retrieval Augmented Generation via Random Projection The authors present a random projection method to protect sensitive datastore embeddings against extraction attacks in retrieval-augmented generation systems. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Evaluates a privacy defense rather than presenting a general benchmark.
AI Consilium: a Benchmark of LLM Clinical Multidisciplinary Second Opinions The authors develop a clinical benchmark evaluating language model pipelines for generating multidisciplinary second opinions on complex medical cases. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Evaluates text-based clinical cases without multimodal image or audio models.
AI Consilium: a Benchmark of LLM Clinical Multidisciplinary Second Opinions The authors develop a clinical benchmark evaluating language model pipelines for generating multidisciplinary second opinions on complex medical cases. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Focuses on clinical diagnostic reasoning without investigating retrieval pipelines.
AI Consilium: a Benchmark of LLM Clinical Multidisciplinary Second Opinions The authors develop a clinical benchmark evaluating language model pipelines for generating multidisciplinary second opinions on complex medical cases. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Applies to clinical diagnosis, not programming or code synthesis.
AI Consilium: a Benchmark of LLM Clinical Multidisciplinary Second Opinions The authors develop a clinical benchmark evaluating language model pipelines for generating multidisciplinary second opinions on complex medical cases. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Evaluates clinical second opinions rather than multi-step tool-using agents.
AI Consilium: a Benchmark of LLM Clinical Multidisciplinary Second Opinions The authors develop a clinical benchmark evaluating language model pipelines for generating multidisciplinary second opinions on complex medical cases. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Presents an evaluation benchmark rather than safety red-teaming or guardrails.
AI Consilium: a Benchmark of LLM Clinical Multidisciplinary Second Opinions The authors develop a clinical benchmark evaluating language model pipelines for generating multidisciplinary second opinions on complex medical cases. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Does not target model distillation, quantization, or inference acceleration.
Knowledge That Is Linear Before Fine-Tuning Generalizes After It The authors study why certain linearly represented facts in language models support generalization to new relations after fine-tuning while other recalled facts do not. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Focuses purely on language model representations, not multimodal data.
Knowledge That Is Linear Before Fine-Tuning Generalizes After It The authors study why certain linearly represented facts in language models support generalization to new relations after fine-tuning while other recalled facts do not. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Does not investigate retrieval systems, document ranking, or grounding.
Knowledge That Is Linear Before Fine-Tuning Generalizes After It The authors study why certain linearly represented facts in language models support generalization to new relations after fine-tuning while other recalled facts do not. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Investigates representation linearity, not code generation or software engineering.
Knowledge That Is Linear Before Fine-Tuning Generalizes After It The authors study why certain linearly represented facts in language models support generalization to new relations after fine-tuning while other recalled facts do not. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Analyzes learning representations without addressing safety, hallucination, or guardrails.
Knowledge That Is Linear Before Fine-Tuning Generalizes After It The authors study why certain linearly represented facts in language models support generalization to new relations after fine-tuning while other recalled facts do not. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Examines representation dynamics without addressing model efficiency or compression.
Knowledge That Is Linear Before Fine-Tuning Generalizes After It The authors study why certain linearly represented facts in language models support generalization to new relations after fine-tuning while other recalled facts do not. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Studies internal representation geometry, not agent planning or tool execution.
Learning to Steer LLM Agents and Multi-Agent Systems The authors propose a learned steering agent that monitors execution context and intervenes to prevent goal drift in tool-using language model systems. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Focuses on an agent steering algorithm rather than benchmark design.
Learning to Steer LLM Agents and Multi-Agent Systems The authors propose a learned steering agent that monitors execution context and intervenes to prevent goal drift in tool-using language model systems. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Does not explore model compression, quantization, or efficient serving.
A team of Gemini 4 Argon agents beats Minecraft The authors report on a team of autonomous Gemini 4 Argon agents collaborating via chat and shared code to beat Minecraft survival mode. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Does not evaluate safety, security, guardrails, or alignment vulnerabilities.
A team of Gemini 4 Argon agents beats Minecraft The authors report on a team of autonomous Gemini 4 Argon agents collaborating via chat and shared code to beat Minecraft survival mode. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Focuses on an empirical multi-agent demonstration rather than benchmark methodology.
AVC-Reward: Event-Centric Audio-Video Consistency Evaluation The authors present AVC-Reward, an event-centric framework to evaluate temporal, narrative, and attribute consistency in generated audio-video content. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on generation fidelity metrics rather than safety or reliability.
AVC-Reward: Event-Centric Audio-Video Consistency Evaluation The authors present AVC-Reward, an event-centric framework to evaluate temporal, narrative, and attribute consistency in generated audio-video content. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on audio-video evaluation, not autonomous planning agents.
AVC-Reward: Event-Centric Audio-Video Consistency Evaluation The authors present AVC-Reward, an event-centric framework to evaluate temporal, narrative, and attribute consistency in generated audio-video content. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:17	Not relevant Focuses on generative multimedia consistency, not industrial visual inspection.
Learning Self-Orchestration in Decentralized Agent Swarms The authors present a market-based post-training framework that allows decentralized agent swarms to learn both action policies and coordination mechanisms without central orchestration. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Focuses on swarm coordination learning rather than retrieval-augmented generation methods.
Learning Self-Orchestration in Decentralized Agent Swarms The authors present a market-based post-training framework that allows decentralized agent swarms to learn both action policies and coordination mechanisms without central orchestration. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Addresses multi-agent swarm coordination rather than software engineering or code generation.
Learning Self-Orchestration in Decentralized Agent Swarms The authors present a market-based post-training framework that allows decentralized agent swarms to learn both action policies and coordination mechanisms without central orchestration. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies agent coordination and bidding rather than AI safety, red-teaming, or guardrails.
Learning Self-Orchestration in Decentralized Agent Swarms The authors present a market-based post-training framework that allows decentralized agent swarms to learn both action policies and coordination mechanisms without central orchestration. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Does not study model compression, pruning, or efficient inference methods.
Learning Self-Orchestration in Decentralized Agent Swarms The authors present a market-based post-training framework that allows decentralized agent swarms to learn both action policies and coordination mechanisms without central orchestration. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Focuses on multi-agent reinforcement learning and coordination rather than multimodal perception.
Learning Self-Orchestration in Decentralized Agent Swarms The authors present a market-based post-training framework that allows decentralized agent swarms to learn both action policies and coordination mechanisms without central orchestration. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes a multi-agent training methodology rather than a benchmark or evaluation dataset.
Twice Upon a Time: Complementary Representations for Time-Series Reasoning The authors propose combining line plots and normalized delay embeddings to provide complementary visual representations for time-series reasoning in language models. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Analyzes time-series reasoning rather than code generation or software engineering.
Twice Upon a Time: Complementary Representations for Time-Series Reasoning The authors propose combining line plots and normalized delay embeddings to provide complementary visual representations for time-series reasoning in language models. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Studies time-series representation in vision-language models, not agent planning or tool use.
Twice Upon a Time: Complementary Representations for Time-Series Reasoning The authors propose combining line plots and normalized delay embeddings to provide complementary visual representations for time-series reasoning in language models. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Does not investigate model quantization, pruning, or serving efficiency.
Twice Upon a Time: Complementary Representations for Time-Series Reasoning The authors propose combining line plots and normalized delay embeddings to provide complementary visual representations for time-series reasoning in language models. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on multimodal time-series modeling rather than deployment safety or hallucination.
Twice Upon a Time: Complementary Representations for Time-Series Reasoning The authors propose combining line plots and normalized delay embeddings to provide complementary visual representations for time-series reasoning in language models. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes a dual-view time-series representation architecture rather than an LLM benchmark.
Twice Upon a Time: Complementary Representations for Time-Series Reasoning The authors propose combining line plots and normalized delay embeddings to provide complementary visual representations for time-series reasoning in language models. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Focuses on visual time-series modeling rather than retrieval-augmented generation systems.
ACEdit: Stable Multi-Turn Image Editing via Asymmetric Conditioning The authors propose ACEdit, a training and distillation scheme that stabilizes image editing models against accumulating errors across multiple editing turns. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on multi-turn image generation and editing rather than language model agents.
ACEdit: Stable Multi-Turn Image Editing via Asymmetric Conditioning The authors propose ACEdit, a training and distillation scheme that stabilizes image editing models against accumulating errors across multiple editing turns. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Addresses image distribution drift in editing rather than AI safety or reliability guardrails.
ACEdit: Stable Multi-Turn Image Editing via Asymmetric Conditioning The authors propose ACEdit, a training and distillation scheme that stabilizes image editing models against accumulating errors across multiple editing turns. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:17	Not relevant Deals with image editing rather than industrial inspection or manufacturing defect detection.
ACEdit: Stable Multi-Turn Image Editing via Asymmetric Conditioning The authors propose ACEdit, a training and distillation scheme that stabilizes image editing models against accumulating errors across multiple editing turns. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Introduces a training scheme for image editing, not an LLM benchmark.
Designing Transformers for Continuous Diffusion Language Models The authors introduce LangDiT, an adapted transformer architecture designed specifically for continuous diffusion language models. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes a model architecture rather than an evaluation framework or benchmark.
Designing Transformers for Continuous Diffusion Language Models The authors introduce LangDiT, an adapted transformer architecture designed specifically for continuous diffusion language models. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on diffusion model perplexity optimization rather than safety, red-teaming, or guardrails.
Designing Transformers for Continuous Diffusion Language Models The authors introduce LangDiT, an adapted transformer architecture designed specifically for continuous diffusion language models. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Develops continuous diffusion language model architectures rather than autonomous planning agents.
KnowGPO: Knowledge-Grouped Policy Optimization for Search Agents The authors propose KnowGPO, a reinforcement learning framework that assigns credit to deep search agents based on acquired knowledge conditions rather than trajectories. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Focuses on training policy optimization for search agents rather than evaluation benchmarks.
KnowGPO: Knowledge-Grouped Policy Optimization for Search Agents The authors propose KnowGPO, a reinforcement learning framework that assigns credit to deep search agents based on acquired knowledge conditions rather than trajectories. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies reward modeling for search agents rather than safety guardrails or reliability risks.
Evaluation Awareness Beyond English: Do LLMs Act Differently When Tested in Indic Languages? The authors evaluate how language models alter their refusal behavior when informed they are being monitored in English and five Indic languages. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Evaluates multilingual safety behavior, not code generation or software engineering.
Evaluation Awareness Beyond English: Do LLMs Act Differently When Tested in Indic Languages? The authors evaluate how language models alter their refusal behavior when informed they are being monitored in English and five Indic languages. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Does not cover model compression, pruning, or efficient inference architectures.
Evaluation Awareness Beyond English: Do LLMs Act Differently When Tested in Indic Languages? The authors evaluate how language models alter their refusal behavior when informed they are being monitored in English and five Indic languages. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Studies evaluation awareness in LLMs, not multi-step agent planning or tool execution.
Evaluation Awareness Beyond English: Do LLMs Act Differently When Tested in Indic Languages? The authors evaluate how language models alter their refusal behavior when informed they are being monitored in English and five Indic languages. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Focuses on text evaluation across languages rather than multimodal vision-language models.

2,221–2,280 of 13,754← Previous Next →