Every paper checked — nothing hidden

All papers

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

All17,312 In brief24 More (on-topic)3,333 Not relevant13,753 Not assessed87 Older351 Could not confirm93 Blogs with no paper266 All blogs284

Paper	Topic	Relevance
DST-Net: Decomposed Spatio-Temporal Network for Time Series Forecasting Presents a decomposed network architecture that applies separate graph and convolution modules to trend and seasonal time series components. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Addresses time series decomposition rather than LLM safety, guardrails, or alignment.
DST-Net: Decomposed Spatio-Temporal Network for Time Series Forecasting Presents a decomposed network architecture that applies separate graph and convolution modules to trend and seasonal time series components. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Designs a forecasting network rather than efficient LLM inference or distillation.
When Judges Agree and Are Wrong: Consistency-Accuracy Dissociation as a Blind Spot for Label-Free Evaluation Examines failure modes in label-free evaluation where models reach high consistency across perturbations while generating identical incorrect answers. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Analyzes label-free evaluation methods rather than autonomous planning or agent tool use.
When Judges Agree and Are Wrong: Consistency-Accuracy Dissociation as a Blind Spot for Label-Free Evaluation Examines failure modes in label-free evaluation where models reach high consistency across perturbations while generating identical incorrect answers. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Examines text-based QA datasets rather than multimodal vision or audio processing.
When Judges Agree and Are Wrong: Consistency-Accuracy Dissociation as a Blind Spot for Label-Free Evaluation Examines failure modes in label-free evaluation where models reach high consistency across perturbations while generating identical incorrect answers. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Investigates consistency metrics rather than external document retrieval and grounding.
When Judges Agree and Are Wrong: Consistency-Accuracy Dissociation as a Blind Spot for Label-Free Evaluation Examines failure modes in label-free evaluation where models reach high consistency across perturbations while generating identical incorrect answers. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Focuses on QA evaluation metrics rather than code generation or software engineering.
When Judges Agree and Are Wrong: Consistency-Accuracy Dissociation as a Blind Spot for Label-Free Evaluation Examines failure modes in label-free evaluation where models reach high consistency across perturbations while generating identical incorrect answers. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Studies evaluation metrics rather than model compression, quantization, or serving efficiency.
Stronger Judges Do Not See More: Deployment-Integrity Blindness in LLM Evaluation Introduces a benchmark measuring how output-only model judges fail to detect tampered runtime environments in agent deployments. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Addresses runtime tamper detection in evaluation rather than agent planning or tool use.
OnBodyAI - Why the Next AI Needs a Body Presents an opinion article on wearable on-body language models based on an industrial panel talk. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Discusses wearable AI vision rather than concrete safety, guardrail, or reliability techniques.
OnBodyAI - Why the Next AI Needs a Body Presents an opinion article on wearable on-body language models based on an industrial panel talk. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Presents an industrial talk overview rather than benchmarks or evaluation methodologies.
OnBodyAI - Why the Next AI Needs a Body Presents an opinion article on wearable on-body language models based on an industrial panel talk. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Offers broad vision statements on wearable AI rather than technical agent frameworks.
CleanRM: Reward Modeling from Noisy Feedback with a Few Clean Feedback for RLHF Proposes an optimal transport method using small clean feedback sets to calibrate sample-wise noise for reward model training. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Addresses reward modeling training objectives rather than LLM evaluation or benchmark suites.
CleanRM: Reward Modeling from Noisy Feedback with a Few Clean Feedback for RLHF Proposes an optimal transport method using small clean feedback sets to calibrate sample-wise noise for reward model training. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on preference feedback denoising rather than multi-step autonomous planning agents.
When is a better LLM worth paying for? Stochastic Optimization with Switching Mixed-Fidelity Oracles Investigates cost-aware policies for adaptively switching between language models of varying fidelity in stochastic optimization pipelines. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Examines stochastic optimization algorithms rather than autonomous agent planning or tool execution.
When is a better LLM worth paying for? Stochastic Optimization with Switching Mixed-Fidelity Oracles Investigates cost-aware policies for adaptively switching between language models of varying fidelity in stochastic optimization pipelines. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Addresses cost tradeoffs in optimization rather than safety guardrails or adversarial threats.
When is a better LLM worth paying for? Stochastic Optimization with Switching Mixed-Fidelity Oracles Investigates cost-aware policies for adaptively switching between language models of varying fidelity in stochastic optimization pipelines. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Focuses on optimization theory rather than LLM benchmarking or evaluation frameworks.
CoSTA: Compressed Speech Token Adaptation for Efficient Spoken Dialogue Models Presents a framework that decouples speech generation and synthesis token spaces to speed up spoken dialogue model inference. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on spoken dialogue tokenization rather than planning or tool-using autonomous agents.
CoSTA: Compressed Speech Token Adaptation for Efficient Spoken Dialogue Models Presents a framework that decouples speech generation and synthesis token spaces to speed up spoken dialogue model inference. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Addresses dialogue latency and throughput rather than deployment safety guardrails or jailbreaks.
CoSTA: Compressed Speech Token Adaptation for Efficient Spoken Dialogue Models Presents a framework that decouples speech generation and synthesis token spaces to speed up spoken dialogue model inference. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Deals with speech token adaptation rather than code generation or software tooling.
CoSTA: Compressed Speech Token Adaptation for Efficient Spoken Dialogue Models Presents a framework that decouples speech generation and synthesis token spaces to speed up spoken dialogue model inference. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Tests speech dialogue models rather than proposing new general LLM evaluation standards.
CoSTA: Compressed Speech Token Adaptation for Efficient Spoken Dialogue Models Presents a framework that decouples speech generation and synthesis token spaces to speed up spoken dialogue model inference. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Addresses speech generation latency rather than external document retrieval and grounding.
All-in-One Memory: Process-Rewarded Reinforcement Learning for Memory Agents Introduces a reinforcement learning framework that evaluates memory operations in agents using a shared state and process rewards. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on memory management optimization rather than safety guardrails or red-teaming.
All-in-One Memory: Process-Rewarded Reinforcement Learning for Memory Agents Introduces a reinforcement learning framework that evaluates memory operations in agents using a shared state and process rewards. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Evaluates a memory agent policy rather than proposing new general LLM evaluation benchmarks.
CARE-SD: Explainable Reasoning Architecture for Local SLM Self-Disclosure Analysis in Textual Data Develops a structured explainable reasoning architecture to extract sensitive self-disclosures from cybersecurity interviews using local small language models. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Evaluates a specific text analysis task rather than broader LLM evaluation methodologies.
CARE-SD: Explainable Reasoning Architecture for Local SLM Self-Disclosure Analysis in Textual Data Develops a structured explainable reasoning architecture to extract sensitive self-disclosures from cybersecurity interviews using local small language models. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on interview text analysis rather than deployment guardrails or adversarial robustness.
CARE-SD: Explainable Reasoning Architecture for Local SLM Self-Disclosure Analysis in Textual Data Develops a structured explainable reasoning architecture to extract sensitive self-disclosures from cybersecurity interviews using local small language models. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Operates exclusively on textual interview transcripts without multimodal image or audio data.
CARE-SD: Explainable Reasoning Architecture for Local SLM Self-Disclosure Analysis in Textual Data Develops a structured explainable reasoning architecture to extract sensitive self-disclosures from cybersecurity interviews using local small language models. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Focuses on classification and reasoning modules rather than retrieval-augmented generation systems.
CARE-SD: Explainable Reasoning Architecture for Local SLM Self-Disclosure Analysis in Textual Data Develops a structured explainable reasoning architecture to extract sensitive self-disclosures from cybersecurity interviews using local small language models. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Applies modular reasoning for text analysis rather than multi-step tool-using planning agents.
CARE-SD: Explainable Reasoning Architecture for Local SLM Self-Disclosure Analysis in Textual Data Develops a structured explainable reasoning architecture to extract sensitive self-disclosures from cybersecurity interviews using local small language models. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Analyzes interview transcripts rather than software code generation or repair.
GRADS: Graph-Regulated Agentic Data Synthesis with Controllable Difficulty Proposes a graph-guided synthetic data generation framework with controlled difficulty levels for training multi-hop research agents. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Designs a data synthesis pipeline rather than establishing LLM evaluation methodologies or benchmarks.
GRADS: Graph-Regulated Agentic Data Synthesis with Controllable Difficulty Proposes a graph-guided synthetic data generation framework with controlled difficulty levels for training multi-hop research agents. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on data synthesis rather than deployment safety, jailbreak defense, or robustness.
MemScope: Evolving Agent Memory Architectures through Localized Code Repair Introduces a localized code repair framework that evolves language model memory architectures using fine-grained execution feedback. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Uses benchmarks to test an agent memory architecture rather than benchmarking LLMs themselves.
MemScope: Evolving Agent Memory Architectures through Localized Code Repair Introduces a localized code repair framework that evolves language model memory architectures using fine-grained execution feedback. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on agent task execution memory rather than safety, guardrails, or alignment risks.
Language Models Share a Readout Bottleneck in Working Memory Analyzes working memory error patterns in language models to show that failures stem from readout interference rather than target retrieval. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on working memory mechanics, not autonomous planning or tool use.
Language Models Share a Readout Bottleneck in Working Memory Analyzes working memory error patterns in language models to show that failures stem from readout interference rather than target retrieval. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Examines internal memory attention across models without introducing efficiency or compression techniques.
Language Models Share a Readout Bottleneck in Working Memory Analyzes working memory error patterns in language models to show that failures stem from readout interference rather than target retrieval. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Analyzes cognitive working-memory mechanics rather than evaluating general LLM benchmarks.
Language Models Share a Readout Bottleneck in Working Memory Analyzes working memory error patterns in language models to show that failures stem from readout interference rather than target retrieval. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Investigates N-back memory tasks rather than code generation or software engineering.
Language Models Share a Readout Bottleneck in Working Memory Analyzes working memory error patterns in language models to show that failures stem from readout interference rather than target retrieval. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Studies internal attention mechanisms, not external document retrieval or grounding.
Language Models Share a Readout Bottleneck in Working Memory Analyzes working memory error patterns in language models to show that failures stem from readout interference rather than target retrieval. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Evaluates text-based memory recall rather than multimodal vision or audio processing.
Language Models Share a Readout Bottleneck in Working Memory Analyzes working memory error patterns in language models to show that failures stem from readout interference rather than target retrieval. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Investigates cognitive mechanistic interpretability rather than safety, guardrails, or jailbreaks.
Judging the Judges: Confidence and Disagreement in Rubrics as Rewards Proposes an ensemble framework that weights judge confidence and masks disagreements when training models with rubrics as rewards. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Evaluates text responses, not multimodal inputs.
Judging the Judges: Confidence and Disagreement in Rubrics as Rewards Proposes an ensemble framework that weights judge confidence and masks disagreements when training models with rubrics as rewards. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Does not address model compression or inference speed.
Judging the Judges: Confidence and Disagreement in Rubrics as Rewards Proposes an ensemble framework that weights judge confidence and masks disagreements when training models with rubrics as rewards. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on reward modeling for RL, not agent tool execution.
Judging the Judges: Confidence and Disagreement in Rubrics as Rewards Proposes an ensemble framework that weights judge confidence and masks disagreements when training models with rubrics as rewards. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Focuses on general scientific and medical QA, not software engineering.
Judging the Judges: Confidence and Disagreement in Rubrics as Rewards Proposes an ensemble framework that weights judge confidence and masks disagreements when training models with rubrics as rewards. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Does not study external retrieval or document grounding.
Does Latent Chain-of-Thought Actually Reason? A Single, Unifying Taxonomy for Latent Reasoning Investigates whether continuous latent reasoning tokens genuinely compute answers or merely settle early on predetermined conclusions. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Studies internal chain-of-thought hidden states, not autonomous agents.
Does Latent Chain-of-Thought Actually Reason? A Single, Unifying Taxonomy for Latent Reasoning Investigates whether continuous latent reasoning tokens genuinely compute answers or merely settle early on predetermined conclusions. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies internal reasoning mechanics rather than deployment safety risks.
Mine, Sample, Render: Test-Time Diversity Alignment Of Generative Models Presents a test-time method that extracts structured schemas from corpora to guide generative models toward more diverse outputs. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on generation diversity rather than safety or reliability.
Mine, Sample, Render: Test-Time Diversity Alignment Of Generative Models Presents a test-time method that extracts structured schemas from corpora to guide generative models toward more diverse outputs. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Does not address model compression, quantization, or latency reduction.
Mine, Sample, Render: Test-Time Diversity Alignment Of Generative Models Presents a test-time method that extracts structured schemas from corpora to guide generative models toward more diverse outputs. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Focuses on diversity sampling rather than benchmarking LLMs.
Mine, Sample, Render: Test-Time Diversity Alignment Of Generative Models Presents a test-time method that extracts structured schemas from corpora to guide generative models toward more diverse outputs. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Proposes a test-time sampling procedure, not agent planning.
FedSumUp: User-Auditable Privacy-Aware Representation Sharing for Resource-Constrained Federated Intelligence Proposes a lossy compression framework for federated learning that balances user privacy against model utility on resource-constrained devices. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Presents a federated learning framework, not an evaluation benchmark.
FedSumUp: User-Auditable Privacy-Aware Representation Sharing for Resource-Constrained Federated Intelligence Proposes a lossy compression framework for federated learning that balances user privacy against model utility on resource-constrained devices. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on federated learning compression, not language model agents.
Condition-Contrastive Extrapolation for On-Policy Distillation of Language Models Introduces a condition-contrastive distillation approach to prevent student language models from learning erroneous teacher continuations. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on training distillation algorithms, not safety guardrails.
Condition-Contrastive Extrapolation for On-Policy Distillation of Language Models Introduces a condition-contrastive distillation approach to prevent student language models from learning erroneous teacher continuations. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Evaluates reasoning distillation rather than proposing evaluation frameworks.
Condition-Contrastive Extrapolation for On-Policy Distillation of Language Models Introduces a condition-contrastive distillation approach to prevent student language models from learning erroneous teacher continuations. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on distillation algorithms, not autonomous agentic systems.
Who May Release an Agent's Promise? Standing over Commitments in Multi-Party Chat Studies whether conversational agents correctly enforce authorization rules when users attempt to cancel prior agent commitments to others. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Does not involve retrieval-augmented generation pipelines.
Who May Release an Agent's Promise? Standing over Commitments in Multi-Party Chat Studies whether conversational agents correctly enforce authorization rules when users attempt to cancel prior agent commitments to others. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Does not cover model compression or inference efficiency.
Who May Release an Agent's Promise? Standing over Commitments in Multi-Party Chat Studies whether conversational agents correctly enforce authorization rules when users attempt to cancel prior agent commitments to others. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Studies commitment tracking rather than general LLM benchmarking.
Who May Release an Agent's Promise? Standing over Commitments in Multi-Party Chat Studies whether conversational agents correctly enforce authorization rules when users attempt to cancel prior agent commitments to others. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Focuses on multi-party conversation, not code generation.

1,621–1,680 of 13,753← Previous Next →