Every paper checked — nothing hidden
All papers
Every paper checked in the week of 12–18 Oct 2026 · LLM evaluation and benchmarks — nothing is hidden. Brief papers first.
All3,345In brief3More (on-topic)587Not relevant2,721Not assessed24Older45Could not confirm25Blogs with no paper54All blogs57
| Paper | Topic | Relevance | Feedback |
|---|---|---|---|
| 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. | |
| 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. | |
| 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. | |
| 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. | |
| 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. | |
| 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 | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on GNN feature imputation rather than language model evaluation. | |
| Certifying Almost-Sure Reachability in Unknown MDPs This paper investigates theoretical limits and conditions for verifying reachability policies in unknown Markov decision processes using sample access. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Does not evaluate language models or introduce LLM benchmarks. | |
| H-DDCM: Hadamard Codebooks for Fast Generative Image and Video Compression This paper presents a generative compression framework using implicit randomized Hadamard dictionaries to speed up diffusion-based image and video coding. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Does not propose benchmarks or evaluations for language models. | |
| Fresh at Decision, Stale at Action: Routing New Information to LLM Decisions Already Made This paper introduces a timestamp theory and persistent ledger to route updated information to pending LLM decisions before action. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents a temporal decision ledger rather than an evaluation benchmark. | |
| Mitigating Cross-Modal Hallucinations by Correcting Modality Misrouting in Omnimodal LLMs This paper proposes an inference-time intervention method to detect and correct modality misrouting that causes multimodal hallucinations. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a hallucination mitigation technique rather than a benchmark. | |
| RadixFlow: Filling Execution Bubbles in Prefix-Cached Hybrid LLM Serving This paper introduces a scheduler that decouples prefix matching from state readiness to eliminate execution bubbles in hybrid LLM serving. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Introduces a serving scheduler rather than an evaluation benchmark. | |
| When Do Learned Blocks Preserve Causal Meaning? Block Identifiability and Constructive Causal Abstractions This paper defines conditions under which learned multidimensional representation blocks preserve causal abstractions and high-level concept interventions. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Theoretical analysis of causal abstractions, not an LLM benchmark. | |
| Fan Out or Relay? Ownership-Aware Routing for LLM Agents Querying Siloed Tables This paper shows data ownership determines whether agents should broadcast or relay table queries, and proposes ownership-aware routing. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a routing architecture rather than an evaluation benchmark. | |
| When and How Order-Flow Interventions Work After Dealers Adapt This paper introduces an order-flow allocation framework and evaluation protocol to analyze how intervention rules influence adaptive trading dealers. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Studies order-flow mechanisms rather than LLM benchmarks. | |
| LoOPSD: On-Policy Self-Distillation with Looped Teachers This paper introduces a self-distillation method that searches over repeated layer execution orders to create stronger teacher guidance. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Introduces a distillation method rather than an evaluation benchmark. | |
| Lifting Finite-Sample Gaussian Approximation of Wide Neural Networks to Functional Convergence This paper establishes functional convergence bounds for wide neural networks approximating Gaussian processes on arbitrary domains. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents mathematical convergence proofs, not empirical evaluation frameworks. | |
| Back to the Future: Reverse-Consequence Modeling for Long-Horizon Particle Dynamics This paper proposes predicting the final state first and using diffusion to interpolate long-horizon physical particle trajectories. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents a physical dynamics model, not an LLM benchmark. | |
| Looped Reasoning Finishes Earlier Than You Think This paper proposes a recurrent architecture that adds rotational degrees of freedom to prevent oscillatory bottlenecks in looped reasoning. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a recurrent architecture rather than an evaluation framework. | |
| Neuro-Symbolic Graph Transformers for Hierarchical Intention Recognition from Multimodal Human Activity Sequences This paper presents a neuro-symbolic graph transformer with first-order logic loss to recognize hierarchical human intentions from multimodal activity sequences. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Introduces a computer vision action dataset, not LLM evaluation. | |
| QORH: Learning Query-Specific Ordered Rule Hypergraphs for Inductive Knowledge Graph Link Prediction QORH models multi-relation rules as query-specific hypergraphs with message passing to improve inductive link prediction on unseen knowledge graph entities. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a knowledge graph model, not an LLM benchmark. | |
| A Few More Tokens Ahead: Reinforcing Parallelism in Diffusion LLMs RePara trains diffusion language models using group-relative reinforcement learning on short token blocks to improve math generation speed and accuracy. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a training and decoding technique rather than evaluation. | |
| Reset Early, Reset Often, Eliminate Models SATResetCoP restarts connection-tableau search branches while preserving accumulated clauses in an incremental SAT solver to improve theorem proving efficiency. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates classical SAT-based provers rather than language model capabilities. | |
| TextualRL: Textual Reinforcement Learning for Context Management TextualRL updates agent context prompts using groupwise trajectory comparisons rather than isolated feedback runs to improve policy adaptation without parameter tuning. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a training method, not a general evaluation benchmark. | |
| 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 | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Audits a specific domain generalization technique rather than LLM benchmarks. | |
| 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. | |
| 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. | |
| 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. | |
| 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. | |
| 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 | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates a security vulnerability rather than general benchmark validity. | |
| 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. | |
| 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. | |
| 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. | |
| 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 | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates video world models rather than language model benchmarks. | |
| 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. | |
| 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. | |
| 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. | |
| AssetForge: Asset-Augmented Task Synthesis for Business Automation Agents AssetForge pairs reusable application assets with task rubrics to synthesize executable training tasks and evaluation checks for business automation agents. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focus is synthetic data generation for agent training, not evaluation. | |
| Lightweight but Linguistically Fair? Quantisation Robustness across Bavarian, German, and English The authors evaluate how 8-bit and 4-bit quantization methods affect model accuracy across Bavarian, Standard German, and English benchmarks. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Uses an existing benchmark to study quantization fairness. | |
| Let the Mind Conceive, Let the Hand Achieve: Following Multimodal Instructions for Long-Horizon Robotic Manipulation Orchestra decomposes multimodal instructions into subtasks, tracks step progress, and maintains execution context to improve long-horizon robotic manipulation. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Introduces a robotics benchmark merely to test its specific method. | |
| ABot-ER: Unified Embodied Reasoning through Agentic Data-Model Co-Evolution The authors introduce an embodied reasoning foundation model that unifies perception, planning, and execution using task-conditioned adaptive adapters. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focus is an embodied agent architecture, not evaluating benchmarks. | |
| Urbex: Agentic Spatial Grounding for City-Scale 3D Scenes The authors present an active spatial search framework enabling vision-language agents to ground 3D objects in city-scale environments via interactive tool use. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents a grounding framework, not a general evaluation benchmark. | |
| Delegating Earth science model calibration to knowledge-guided LLM agents The authors introduce a calibration harness using structured domain knowledge to safely delegate Earth science model parameterization to language model agents. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents an agent workflow harness, not an LLM evaluation benchmark. | |
| Graph-free Retrieval with Graph-Modified Embeddings The authors evaluate offline graph-modified embeddings for vector retrieval, improving multi-hop evidence recovery without requiring query-time graph access. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a retrieval method, not an LLM benchmark or evaluation protocol. | |
| Resource-Efficient 3D Shape and Orientation Estimation of Road Users using Roadside RGB Camera and 3D Radar The authors propose a resource-efficient two-stage approach for estimating 3D road user shapes and orientations from roadside cameras and radar. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Computer vision perception model, not an LLM benchmark. | |
| Agaram 2.5B: The First Tamil-First Language Model Trained from Scratch with Tool Calling. A Playbook for Under-Represented Languages The authors introduce a 2.5-billion-parameter bilingual Tamil-English language model trained from scratch with tool-calling capabilities. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Training playbook for low-resource languages, not an evaluation methodology. | |
| FedGraph: Defending Federated Large Language Model Fine-Tuning Against Backdoor Attacks via Graph-Based Aggregation The authors present a graph-based aggregation method to defend federated language model fine-tuning against backdoor attacks by identifying update consistency patterns. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents a defense framework, not an LLM benchmark or evaluation study. | |
| STAMP as a topology-aware molecular language for fragment-level learning and editing The authors introduce a topology-aware molecular representation language that enables fragment-level representation learning and precise structural edits. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on molecular editing representations, not LLM evaluation. | |
| The Singularity Has a Distribution: Exact Timing Laws and Detection Limits for a Stochastic Intelligence Explosion The authors formulate a stochastic model to analyze the timing distributions and detection limits of potential recursive artificial intelligence explosions. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Theoretical timing analysis rather than LLM benchmarking or evaluation. | |
| Thinker-WM: A Unified World Model with Shared Multimodal Representations for Embodied Intelligence The authors propose an embodied world model that unifies RGB video, metric depth, and robot actions into a shared dynamic representation. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a world model architecture, not a benchmark or evaluation method. | |
| Learning Perturbation-Specific Representations through Conditional Energy Prior for Cellular Response Prediction The authors present an energy-based prior for cellular perturbation models to better distinguish specific therapeutic signals from general biological background noise. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Domain-specific biological predictive model, not an LLM benchmark. | |
| Readout Scaling Shapes Sparse Feature Recovery The authors analyze readout scaling in sparse autoencoders and propose canonicalized coordinates to improve the recovery of underlying generative factors. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Mechanistic interpretability study, not LLM benchmark evaluation. | |
| MemoryVLN: From Ego-Aware Memory Representation to Cross-Context Memory Reuse in Vision-Language Navigation The authors propose a memory-driven vision-language navigation system using temporal compression and geometric-semantic representations to support long-horizon trajectory tasks. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Introduces a navigation architecture, not an evaluation framework or benchmark. | |
| ReTerm: Recovery Training for Self-Improving Terminal Agents The authors present a training framework and dataset that enables terminal coding agents to self-improve by recovering from execution errors. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Introduces a training method, not a benchmark or evaluation study. | |
| Task-evoked eye movements as candidate digital biomarkers of depression severity in adolescents and young adults The authors analyze task-evoked eye movements in adolescents to identify objective behavioral markers corresponding to depression severity scores. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Not an evaluation or benchmark of language models. | |
| The Smart Buildings Challenge: Learning Foundation Models for Real-World Optimization at Scale The authors introduce a challenge benchmark to develop foundation models for building HVAC energy optimization that transfer across different physical structures. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents a challenge for building HVAC control, not LLM evaluation. | |
| The Geometry of Language Model Logits: Generic Structure and Learned Departures The paper analyzes the geometric distribution of language model output logits, identifying a consistent log-linear relationship that emerges during pretraining. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Investigates foundational logit geometry rather than evaluation benchmarks. | |
| Lost in Implicit Routing: How LLM Agents Fail to Load Relevant Skills at Scale The paper shows that large language model agents fail to retrieve relevant skills when large numbers of irrelevant skills populate the metadata library. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Studies skill-loading failure modes rather than proposing an LLM benchmark. | |
| APTree: Extended Behavior Trees for Adaptive and Scalable Automated Robotic Task Planning in Construction The paper presents an extended behavior tree framework that automates modular robot task planning and runtime replanning for off-site construction fabrication. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents a robotic execution architecture, not an LLM benchmark. |
2,641–2,700 of 3,345← PreviousNext →