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
A Controlled Evaluation of Curvature Inspired Scalar Feature Scaling for Domain Generalization This paper audits a specific scalar feature scaling method to test whether claimed curvature benefits hold for domain generalization benchmarks. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Investigates feature scaling for generalization, not model compression or efficiency.
Holding Agents Accountable with Verifiable Agent Insurance The authors propose verifiable agent insurance using smart-contract escrow to financially compensate users when agents violate programmatically logged behavioral promises. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Proposes an enforcement framework rather than general LLM evaluation.
Consider, Then Choose: Order-Invariant Decisions in System One Models Runoff eliminates option-order bias in multiple-choice model predictions by scoring items independently before running balanced candidate comparisons. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Investigates multiple-choice text evaluation, not multimodal models.
Consider, Then Choose: Order-Invariant Decisions in System One Models Runoff eliminates option-order bias in multiple-choice model predictions by scoring items independently before running balanced candidate comparisons. Preprint	LLMs for code via arxiv · 15 Oct, 06:17	Not relevant Studies option order invariance, not software code generation.
Consider, Then Choose: Order-Invariant Decisions in System One Models Runoff eliminates option-order bias in multiple-choice model predictions by scoring items independently before running balanced candidate comparisons. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Studies decision invariance, not quantization or model distillation.
Consider, Then Choose: Order-Invariant Decisions in System One Models Runoff eliminates option-order bias in multiple-choice model predictions by scoring items independently before running balanced candidate comparisons. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17	Not relevant Focuses on multiple-choice decision bias, not document retrieval.
Consider, Then Choose: Order-Invariant Decisions in System One Models Runoff eliminates option-order bias in multiple-choice model predictions by scoring items independently before running balanced candidate comparisons. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Addresses option-order bias in question answering, not agent planning.
Heading Hierarchy Reconstruction for Complex and Long Documents with Level-Wise Multimodal Memory The authors present a framework using level-wise multimodal memory to reconstruct coherent heading hierarchies across long, multi-page documents. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focus is document layout understanding, not AI deployment safety.
Heading Hierarchy Reconstruction for Complex and Long Documents with Level-Wise Multimodal Memory The authors present a framework using level-wise multimodal memory to reconstruct coherent heading hierarchies across long, multi-page documents. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on document hierarchy reconstruction, not autonomous agents.
Heading Hierarchy Reconstruction for Complex and Long Documents with Level-Wise Multimodal Memory The authors present a framework using level-wise multimodal memory to reconstruct coherent heading hierarchies across long, multi-page documents. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Focus is multimodal document parsing architecture, not evaluation methodology.
Heading Hierarchy Reconstruction for Complex and Long Documents with Level-Wise Multimodal Memory The authors present a framework using level-wise multimodal memory to reconstruct coherent heading hierarchies across long, multi-page documents. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:17	Not relevant Addresses document parsing, not industrial visual quality control.
Beyond Answer Sharing: TrueMAD for Targeted Peer Engagement in Multi-Agent Debate TrueMAD introduces one-to-one peer feedback between debate rounds to help language model agents resolve conflicting reasoning before final revision. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Evaluates on benchmarks but focuses on multi-agent debate architecture.
Beyond Answer Sharing: TrueMAD for Targeted Peer Engagement in Multi-Agent Debate TrueMAD introduces one-to-one peer feedback between debate rounds to help language model agents resolve conflicting reasoning before final revision. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focus is multi-agent reasoning accuracy, not safety guardrails.
LiFUSE: Frequency-Spatial Unified Self-Supervised Pretraining for Liver Disease Analysis on Non-contrast CT LiFUSE uses frequency-spatial self-supervised pretraining on non-contrast CT scans to improve representation learning for liver disease detection. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Addresses self-supervised pretraining, not model compression or efficiency.
LiFUSE: Frequency-Spatial Unified Self-Supervised Pretraining for Liver Disease Analysis on Non-contrast CT LiFUSE uses frequency-spatial self-supervised pretraining on non-contrast CT scans to improve representation learning for liver disease detection. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Does not investigate AI safety guardrails or prompt risks.
LiFUSE: Frequency-Spatial Unified Self-Supervised Pretraining for Liver Disease Analysis on Non-contrast CT LiFUSE uses frequency-spatial self-supervised pretraining on non-contrast CT scans to improve representation learning for liver disease detection. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:17	Not relevant Focuses on medical liver CT imaging, not industrial inspection.
LiFUSE: Frequency-Spatial Unified Self-Supervised Pretraining for Liver Disease Analysis on Non-contrast CT LiFUSE uses frequency-spatial self-supervised pretraining on non-contrast CT scans to improve representation learning for liver disease detection. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Proposes medical self-supervised pretraining, not LLM evaluation benchmarks.
LiFUSE: Frequency-Spatial Unified Self-Supervised Pretraining for Liver Disease Analysis on Non-contrast CT LiFUSE uses frequency-spatial self-supervised pretraining on non-contrast CT scans to improve representation learning for liver disease detection. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on medical CT pretraining, not LLM agents.
LiFUSE: Frequency-Spatial Unified Self-Supervised Pretraining for Liver Disease Analysis on Non-contrast CT LiFUSE uses frequency-spatial self-supervised pretraining on non-contrast CT scans to improve representation learning for liver disease detection. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Focuses on medical 3D CT analysis, not multimodal VLMs.
FALLNEST: Dual-Domain Dataset Condensation of Imbalanced IMU Data for Wearable Fall Detection FALLNEST condenses large imbalanced wearable motion sensor datasets into small balanced synthetic sets for training fall detection models. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Deals with sensor data condensation, not language model agents.
FALLNEST: Dual-Domain Dataset Condensation of Imbalanced IMU Data for Wearable Fall Detection FALLNEST condenses large imbalanced wearable motion sensor datasets into small balanced synthetic sets for training fall detection models. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:17	Not relevant Focuses on wearable sensor fall detection, not industrial vision.
FALLNEST: Dual-Domain Dataset Condensation of Imbalanced IMU Data for Wearable Fall Detection FALLNEST condenses large imbalanced wearable motion sensor datasets into small balanced synthetic sets for training fall detection models. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Processes IMU time-series sensor signals, not vision-language models.
FALLNEST: Dual-Domain Dataset Condensation of Imbalanced IMU Data for Wearable Fall Detection FALLNEST condenses large imbalanced wearable motion sensor datasets into small balanced synthetic sets for training fall detection models. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focus is wearable sensor classification, not AI deployment safety.
FALLNEST: Dual-Domain Dataset Condensation of Imbalanced IMU Data for Wearable Fall Detection FALLNEST condenses large imbalanced wearable motion sensor datasets into small balanced synthetic sets for training fall detection models. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Focus is wearable dataset condensation, not LLM evaluation.
Encrypted Skills Can Become New Security Exploits If AI Agents Know How to Decipher The authors demonstrate that encoding malicious instructions in agent skills bypasses scanners and causes models to execute harmful actions upon decoding. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17	Not relevant Studies prompt injection in agent skills, not document retrieval.
Encrypted Skills Can Become New Security Exploits If AI Agents Know How to Decipher The authors demonstrate that encoding malicious instructions in agent skills bypasses scanners and causes models to execute harmful actions upon decoding. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Evaluates a security vulnerability rather than general benchmark validity.
Encrypted Skills Can Become New Security Exploits If AI Agents Know How to Decipher The authors demonstrate that encoding malicious instructions in agent skills bypasses scanners and causes models to execute harmful actions upon decoding. Preprint	LLMs for code via arxiv · 15 Oct, 06:17	Not relevant Focus is agent security vulnerabilities, not coding assistants.
Encrypted Skills Can Become New Security Exploits If AI Agents Know How to Decipher The authors demonstrate that encoding malicious instructions in agent skills bypasses scanners and causes models to execute harmful actions upon decoding. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Does not address model compression, pruning, or efficient inference.
Encrypted Skills Can Become New Security Exploits If AI Agents Know How to Decipher The authors demonstrate that encoding malicious instructions in agent skills bypasses scanners and causes models to execute harmful actions upon decoding. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Investigates text-based agent prompt injection, not multimodal models.
JevGPT: Thinking Fast and Slow in Autonomous Research JevGPT combines fast local evidence assessments with deliberative planning to improve autonomous research agents operating under fixed retrieval budgets. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Processes text documents and question answering, not multimodal data.
JevGPT: Thinking Fast and Slow in Autonomous Research JevGPT combines fast local evidence assessments with deliberative planning to improve autonomous research agents operating under fixed retrieval budgets. Preprint	LLMs for code via arxiv · 15 Oct, 06:17	Not relevant Applies to autonomous research and QA, not software engineering.
JevGPT: Thinking Fast and Slow in Autonomous Research JevGPT combines fast local evidence assessments with deliberative planning to improve autonomous research agents operating under fixed retrieval budgets. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Introduces an autonomous research system, not an evaluation benchmark.
JevGPT: Thinking Fast and Slow in Autonomous Research JevGPT combines fast local evidence assessments with deliberative planning to improve autonomous research agents operating under fixed retrieval budgets. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focus is autonomous research efficiency, not safety or guardrails.
JevGPT: Thinking Fast and Slow in Autonomous Research JevGPT combines fast local evidence assessments with deliberative planning to improve autonomous research agents operating under fixed retrieval budgets. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Focus is dual-system agent coordination, not model compression.
Which Optimizer Leaks Most? Optimizer Implicit Bias as a Training-Data Privacy Controller The authors analyze how different optimizers leave distinct implicit bias fingerprints that influence training data reconstruction from model checkpoints. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Focuses on training data reconstruction, not model evaluation.
Which Optimizer Leaks Most? Optimizer Implicit Bias as a Training-Data Privacy Controller The authors analyze how different optimizers leave distinct implicit bias fingerprints that influence training data reconstruction from model checkpoints. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Studies privacy leakage from optimizers, not agent planning.
Which Optimizer Leaks Most? Optimizer Implicit Bias as a Training-Data Privacy Controller The authors analyze how different optimizers leave distinct implicit bias fingerprints that influence training data reconstruction from model checkpoints. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Studies optimizer privacy leakage, not model efficiency or compression.
Output Discipline Sets the Floor: A Falsification-Guided, Model-Agnostic Agent Protocol for Auditable Hypothesis-Driven Decision Making The paper introduces a falsification-guided agent protocol requiring systematic hypothesis elimination to improve decision-making consistency on complex diagnostic tasks. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Focus is agent decision framework, not evaluation methodology.
A dataset of formalised theorem statements from the Annals of Mathematics AnnalsChallenge provides 50 formalized mathematics theorem statements in Lean 4 drawn from papers published in the Annals of Mathematics. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focus is mathematical formalisation benchmarking, not deployment safety.
A dataset of formalised theorem statements from the Annals of Mathematics AnnalsChallenge provides 50 formalized mathematics theorem statements in Lean 4 drawn from papers published in the Annals of Mathematics. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Provides a formal mathematics dataset, not an agent system.
Efficient Safety Guardrails through Latent Dual-Hypothesis Reasoning with On-Demand Trace Reconstruction LR-Guard distills explicit safety reasoning into compact latent representations to speed up guardrail classification while enabling on-demand trace reconstruction. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focus is safety guardrail inference, not planning agents.
Efficient Safety Guardrails through Latent Dual-Hypothesis Reasoning with On-Demand Trace Reconstruction LR-Guard distills explicit safety reasoning into compact latent representations to speed up guardrail classification while enabling on-demand trace reconstruction. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Introduces a guardrail model rather than general benchmark evaluation.
AffectWorld: Benchmarking Emotion-Driven Future World Prediction AffectWorld benchmarks world models on predicting emotion-driven human actions and subsequent environment changes across annotated video samples. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focus is video world prediction, not safety or guardrails.
AffectWorld: Benchmarking Emotion-Driven Future World Prediction AffectWorld benchmarks world models on predicting emotion-driven human actions and subsequent environment changes across annotated video samples. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:17	Not relevant Benchmark evaluates emotion-driven world prediction, not industrial inspection.
AffectWorld: Benchmarking Emotion-Driven Future World Prediction AffectWorld benchmarks world models on predicting emotion-driven human actions and subsequent environment changes across annotated video samples. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Evaluates video world models rather than language model benchmarks.
AffectWorld: Benchmarking Emotion-Driven Future World Prediction AffectWorld benchmarks world models on predicting emotion-driven human actions and subsequent environment changes across annotated video samples. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Evaluates video world models, not tool-using language model agents.
SimInter: Evidence-Grounded User Simulation with Interview-Based Profile Enrichment for Conversational Recommender Systems SimInter simulates conversational recommendation users by conducting synthetic interviews between language models to enrich implicit behavioral profiles. Preprint	Small and efficient models via arxiv · 15 Oct, 06:17	Not relevant Does not address model compression, quantization, or efficient serving.
SimInter: Evidence-Grounded User Simulation with Interview-Based Profile Enrichment for Conversational Recommender Systems SimInter simulates conversational recommendation users by conducting synthetic interviews between language models to enrich implicit behavioral profiles. Preprint	LLMs for code via arxiv · 15 Oct, 06:17	Not relevant Does not involve programming, code generation, or software engineering.
SimInter: Evidence-Grounded User Simulation with Interview-Based Profile Enrichment for Conversational Recommender Systems SimInter simulates conversational recommendation users by conducting synthetic interviews between language models to enrich implicit behavioral profiles. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Introduces a user simulation method, not an LLM benchmark.
SimInter: Evidence-Grounded User Simulation with Interview-Based Profile Enrichment for Conversational Recommender Systems SimInter simulates conversational recommendation users by conducting synthetic interviews between language models to enrich implicit behavioral profiles. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Contains an audit agent but main focus is user simulation.
SimInter: Evidence-Grounded User Simulation with Interview-Based Profile Enrichment for Conversational Recommender Systems SimInter simulates conversational recommendation users by conducting synthetic interviews between language models to enrich implicit behavioral profiles. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:17	Not relevant Focuses on conversational recommendation simulation, not document retrieval.
SimInter: Evidence-Grounded User Simulation with Interview-Based Profile Enrichment for Conversational Recommender Systems SimInter simulates conversational recommendation users by conducting synthetic interviews between language models to enrich implicit behavioral profiles. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Works on text-based conversational systems, not multimodal inputs.
Complex materials engineering via Provenance Action Graphs: potential, structure and demonstration This paper examines using provenance reaction action graphs and machine learning to encode multi-step material synthesis pathways from experimental history. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Does not evaluate language models or benchmarking methodology.
Complex materials engineering via Provenance Action Graphs: potential, structure and demonstration This paper examines using provenance reaction action graphs and machine learning to encode multi-step material synthesis pathways from experimental history. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Focuses on reaction graphs for materials science, not LLM agents.
Complex materials engineering via Provenance Action Graphs: potential, structure and demonstration This paper examines using provenance reaction action graphs and machine learning to encode multi-step material synthesis pathways from experimental history. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Focuses on materials synthesis graphs, not AI safety.
ShoeFit: A New Dataset and Dual-image-stream DiT Framework for Virtual Footwear Try-On ShoeFit presents a multi-view dataset and a dual-stream diffusion framework for consistent virtual footwear try-on synthesis. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:17	Not relevant Focuses on virtual footwear try-on, not industrial inspection.
ShoeFit: A New Dataset and Dual-image-stream DiT Framework for Virtual Footwear Try-On ShoeFit presents a multi-view dataset and a dual-stream diffusion framework for consistent virtual footwear try-on synthesis. Preprint	Multimodal models via arxiv · 15 Oct, 06:17	Not relevant Focus is generative image diffusion, not multimodal reasoning models.
ShoeFit: A New Dataset and Dual-image-stream DiT Framework for Virtual Footwear Try-On ShoeFit presents a multi-view dataset and a dual-stream diffusion framework for consistent virtual footwear try-on synthesis. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17	Not relevant Releases a dataset for footwear try-on, not LLM benchmarks.
ShoeFit: A New Dataset and Dual-image-stream DiT Framework for Virtual Footwear Try-On ShoeFit presents a multi-view dataset and a dual-stream diffusion framework for consistent virtual footwear try-on synthesis. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:17	Not relevant Does not focus on AI safety, alignment, or guardrails.
ShoeFit: A New Dataset and Dual-image-stream DiT Framework for Virtual Footwear Try-On ShoeFit presents a multi-view dataset and a dual-stream diffusion framework for consistent virtual footwear try-on synthesis. Preprint	AI agents via arxiv · 15 Oct, 06:17	Not relevant Covers image generation for try-on, not LLM agents.

10,441–10,500 of 13,754← Previous Next →