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 |
|---|---|---|---|
| Co-adaptive Rule Routing and Constraining for Structured Knowledge Retrieval The paper presents a framework that jointly learns query rule assignment and rule-constrained evidence retrieval over structured knowledge bases. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Develops a retrieval method rather than a general evaluation benchmark. | |
| Diffusion-Based Auctions for Multi-Winner Banner Ads The paper proposes a multi-winner banner ad auction mechanism using diffusion models guided by bids and text descriptions to generate ads. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Does not benchmark or evaluate language models. | |
| RES-MR: Risk-Aware Reasoning for Explainable and Safe Medication Recommendation The paper introduces a two-stage reasoning framework to generate personalized, safe, and explainable medication recommendations using language models. Peer-reviewed | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on a specialized clinical recommendation framework, not general LLM benchmarking. | |
| AphaVoice: Text-to-Speech Model for Aphasia Patient Simulation The paper develops a text-to-speech model that simulates realistic clinical dysfluency in aphasia speech across multiple speakers for training healthcare professionals. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a specialized audio generator rather than benchmark evaluations. | |
| AlignPoison: Conditional Targeted Poisoning of Generative Recommenders via Decoding-Aligned Prefixes The paper demonstrates that generative recommenders are vulnerable to targeted poisoning attacks using fewer than 1% malicious synthetic interactions. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Examines adversarial poisoning rather than LLM evaluation benchmarks. | |
| HiPPT: Physiology-Grounded Wearable PPG Representation Learning via Multi-Scope Text Alignment The paper presents a pretraining framework aligning wearable PPG signals with text across both individual pulse morphologies and broader demographic contexts. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on sensor representation learning rather than LLM benchmarks. | |
| Reels-RSI: Program-Level Self-Improvement for Code-Rendered Explainer Videos The paper introduces an open-source agent framework that improves code-rendered explainer videos through program-level debugging, static rules, and unit testing. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on a self-improving coding agent system rather than LLM benchmarking. | |
| Rethinking Activation Functions: Reducing Bias Can Improve Learning The paper examines positive activation bias in neural networks, showing that reducing this bias slows initial training but improves final model performance. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Studies optimization dynamics rather than benchmark evaluation frameworks. | |
| MontageFlow: Physically Valid High Fidelity EEG Generation The paper introduces a flow-matching model that enforces exact linear identity constraints among EEG channels to generate physically valid clinical signals. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on generative signal modeling rather than LLM benchmarking. | |
| Compositional Concepts Erasing: Disentangling Object-Style via Domain Manifold Unlearning The paper presents a feature-proxy method to selectively erase combined object-style concepts from diffusion models without degrading general generative performance. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on machine unlearning methods rather than benchmark evaluation. | |
| ScienceBuddy-PV: A Verifier-Governed Recursive Improvement Protocol for Panvascular Intervention Research The paper specifies a verifier-governed protocol for panvascular research and benchmarks a baseline model on intervention imaging tasks. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates medical computer vision components rather than LLM benchmarks. | |
| Energy Ring: Stabilizing Feature Energy for Reliable Test-Time Discovery The paper proposes an energy-anchoring technique to reliably separate known and novel classes during continual test-time model adaptation. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents a representation discovery method rather than benchmark evaluation. | |
| Who to Trust in Multi-Robot Systems: Joint Robot and Object Hypothesis Trust Inference via Neural-Symbolic Cross-Validation The paper presents a neural-symbolic method for multi-robot systems to evaluate robot and object hypothesis trust purely via inter-robot observation consistency. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a Bayesian trust inference framework rather than LLM benchmark evaluation. | |
| GG-Nav: Goal-oriented Grounding Chain-of-Thought Elicits Reasoning in Navigation Foundation Models The paper proposes a structured chain-of-thought grounding method to track visual evidence across video frames for goal-oriented embodied navigation. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a navigation framework rather than general LLM evaluation methodology. | |
| A Self-Defining Operator for Distributed Applications The paper presents an autonomous agent architecture that distills distributed system incident resolutions into reusable detectors and operational playbooks. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on an operational troubleshooting agent rather than benchmark design. | |
| Active Neural Matter: A Finite-Window Response Framework for Physical Systems and Distributed AI Workflows The paper introduces a finite-window theoretical framework to determine what historical state information dynamical systems and distributed AI workflows must retain. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Investigates finite-window state response theory rather than LLM benchmarks. | |
| LASER: Defeasible Certificate Search for Long-Range Protocol Dependencies The paper presents a certificate-based search method paired with a symbolic checker to resolve complex, distributed dependencies in protocol specifications. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a formal verification search method rather than evaluating LLMs. | |
| Learning after Execution with Action-Specific Credit The paper proposes a reinforcement learning method that assigns distinct credit to initial code generation and subsequent post-execution correction steps. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on reinforcement learning credit assignment, not benchmark evaluation methodologies. | |
| CB-Agent: Concept-Based In-Context Learning for Sequential Action Selection The paper presents an in-context learning method that extracts compact concept vectors from demonstrations to guide sequential action choices for agents. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes an in-context learning agent method rather than benchmark evaluation. | |
| Iterate on an LLM, Serve on an SLM The paper proposes prototyping agent workflows with large language models and distilling converged live production interactions into smaller models. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on deployment workflows rather than benchmarking or evaluation metrics. | |
| Decoupling Knowledge Forgetting and Refusal Learning in MLLM Unlearning The paper proposes a distillation framework to remove specific multimodal knowledge without causing hallucinations or leaving residual queryable information. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Introduces an unlearning framework rather than general LLM evaluation benchmarks. | |
| Limitations on Autoregressive Token Generation via Sequential Sampling The paper analyzes theoretical computational limits of sequential sampling circuits in autoregressive language model token generation. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Theoretical circuit complexity study, not empirical LLM evaluation or benchmarking. | |
| EarthAlign: Learning Language-Aligned Representations of Earth System Data The paper releases a multimodal dataset and dual-encoder model to align natural language descriptions with global Earth system meteorological data. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents a domain data alignment pipeline, not general LLM benchmark evaluation. | |
| From Existing Methods to New Research Ideas: Two Perspectives on Recursive Self-Improvement The paper presents a unified taxonomy and design space organizing recursive self-improvement methods for language model agents. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant A taxonomy and survey on self-improvement, not an evaluation benchmark study. | |
| PEARL: Process-Experience Aware Reinforcement Learning for Multi-Turn Agents The paper introduces a reinforcement learning approach for multi-turn agents that identifies recurring decision patterns to improve intermediate step credit assignment. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a reinforcement learning algorithm rather than benchmark or evaluation methodology. | |
| AegisVox: Factorized Privacy and Clinical Preservation for Long-Form Speech The paper evaluates a framework that replaces spoken patient identifiers and alters voice identity while preserving clinical conversational evidence. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates speech redaction rather than language model benchmarks or LLM-as-judge. | |
| Shaping Latent Mixture-of-Experts: From Expert Geometry to Layerwise Capacity Allocation The paper studies expert width and dimension balance in latent mixture-of-experts architectures to optimize parameter capacity across model layers. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Addresses model architecture scaling rather than evaluation methodologies or benchmarks. | |
| Prospective Forcing: Learning to Look Ahead for Autoregressive Long Video Diffusion The paper introduces a multi-head prediction method that injects future long-range supervision to improve autoregressive video generation consistency. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on video diffusion architecture rather than language model evaluation. | |
| RASSG: Repair-Aware Semantic Survival Gating for Streaming Tool Execution The paper proposes an execution-gating method to safely launch and cancel tool calls early during live speech recognition before user self-corrections. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on execution gating algorithms rather than establishing general LLM benchmarks. | |
| A Hybrid Kinematic GCN and Elastic Transformer for Efficient Indian Sign Language Recognition The paper introduces a hybrid kinematic graph neural network and elastic transformer framework for efficient Indian sign language recognition from video. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on sign language recognition architectures, not LLM benchmark evaluation. | |
| DexCast: Learning Real-World Dexterous Control from Human Videos The paper presents a framework to learn dexterous robotic hand control directly from human monocular video demonstrations without teleoperation. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Does not propose or analyze language model benchmarks or evaluation methods. | |
| AcrossVAM2.0: Conflict-Aware Point Supervision for Parameter-Efficient Video Diffusion Adaptation This paper presents a parameter-efficient video diffusion adaptation method combining point tracking with gradient conflict resolution to preserve visual priors. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on parameter adaptation for video diffusion, not general LLM benchmarking. | |
| AlignAudio: A Unified Dual-Alignment Framework for Noise-Robust Speech-Conditioned Generation This paper proposes a dual-alignment framework to keep speech-conditioned audio generation robust and consistent across clean and noisy acoustic environments. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on audio generation modeling rather than LLM evaluation benchmarks. | |
| Joint Discrete and SE(3) Diffusion for Protein Interface Design This paper presents a multimodal diffusion framework that jointly designs protein sequences and physical structures to ensure interface complementarity. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Dataset introduction is specific to protein binder evaluation, not LLM benchmarking. | |
| Towards Merging-conditional Backdoor Unalignment against Large Language Models This paper demonstrates a security risk where dormant backdoors trigger safety failures only after merging models that passed individual safety audits. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on backdoor attack mechanics rather than general benchmark evaluation methodology. | |
| Lingdan-V2: Large language models for clinically grounded reasoning in traditional Chinese medicine This paper introduces an aligned language model family trained with structured reinforcement learning for clinical reasoning and traditional Chinese medicine prescribing. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Benchmark evaluation is purely secondary to developing domain-specific clinical models. | |
| Vela: User Behavior as Supervision Sparks LLM Intelligence This paper presents a framework that converts long-term real user behavior records into supervisory tasks to train personalization-focused language models. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents a training framework rather than an LLM evaluation benchmark. | |
| Relational Semantic Grounding for Test-Time Discovery This paper proposes a framework using language model attributes to ground and label novel visual classes discovered dynamically at test time. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents a class discovery method rather than LLM benchmarking. | |
| Integrating AIRL-Based Reward Estimation into a Control Method Using Fuzzy Inference and Policy Gradient This paper integrates adversarial inverse reinforcement learning into a fuzzy control framework to estimate reward functions directly from demonstration data. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes a fuzzy control method, not an LLM benchmark. | |
| MHWA: Multi-timescale Hierarchical World-Action Model This paper introduces a hierarchical multi-timescale world model to reduce compounding rollout errors and drift in sequential decision-making systems. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents an RL architecture rather than language model evaluation. | |
| Self-improving data synthesis can scale embedding models across domains and tasks This paper introduces a self-improving synthetic data generation pipeline to train embedding models across diverse retrieval domains without extensive manual annotation. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Introduces a data synthesis method rather than an evaluation benchmark. | |
| Think in Strategies, Act with Tools: Traceable Agentic Optimization for Combinatorial Problems This paper presents a framework that uses history tracking and problem-specific tools to help language agents solve complex combinatorial optimization tasks. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Presents an optimization agent framework rather than an evaluation benchmark. | |
| PTGPO: Prefix Tree Group Policy Optimization This paper proposes a critic-free reinforcement learning algorithm using prefix trees to assign step-level credit for interactive multi-step tasks. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes an RL training algorithm, not an LLM benchmark. | |
| SAGE: Pushing 2-Bit Vector-Quantized LLMs toward Real-World Usability This paper introduces energy-based channel grouping and bit allocation to preserve accuracy when quantizing large language models to two bits. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates low-bit quantization schemes, not LLM benchmark methodologies. | |
| Equal-information evaluation of credit assignment in modular financial reinforcement learning This paper proposes an evaluation framework to test whether credit assignment techniques in modular financial reinforcement learning truly improve economic decisions. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates RL credit assignment methods, not general LLM benchmarking. | |
| GateSight: Recurrent Memory Anticipates Attention Memory for Query-Agnostic KV Eviction in Hybrid LLMs This paper shows recurrent layers can predict future attention importance, enabling memory eviction in hybrid language models before queries arrive. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on memory management techniques, not evaluation benchmarks. | |
| CUSE: Fine-Grained Multimodal Unlearning with Counterfactual Guidance and On-Policy Self-Distillation This paper introduces a multimodal unlearning method using counterfactual inputs and on-policy self-distillation to remove sensitive data without damaging general knowledge. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes an unlearning method rather than benchmarking or evaluating LLMs. | |
| Prior-Fitted and Frozen Scorers for Multimodal Contrastive Learning This paper investigates using frozen, pre-trained Transformer scorers instead of fixed mathematical formulas for multimodal contrastive representation learning. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Examines representation learning scoring functions, not LLM benchmark evaluation. | |
| How Should a Camera-First AI Agent Work? This paper proposes a two-deadline architecture for visual agents to rapidly capture fleeting visual targets before slower semantic reasoning completes. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Proposes an agent architecture rather than an LLM evaluation benchmark. | |
| ProgAgent: Learning Tool Orchestration through Programmatic Training This paper presents a training framework that teaches language models to orchestrate tools through executable code instead of sequential prompt accumulation. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Introduces training methods rather than general LLM evaluation standards. | |
| Training a Compact, Strong, and Realistic Text-to-Speech Model This paper introduces a compact text-to-speech model trained with a realism critic to improve prosodic naturalness during zero-shot voice cloning. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Develops a speech synthesis model rather than evaluating language models. | |
| LYNX: Analytical Program Synthesis via Grounding and Closed-Loop Refinement This paper presents a multi-agent framework that translates natural language into verified, executable data analytical programs using schema-grounded refinement. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates analytical synthesis systems rather than proposing LLM benchmark methodology. | |
| Counterfactual Regret for Reflective Agents with Lean-Verified Guarantees This paper establishes mathematical conditions to guarantee long-term performance and low counterfactual regret for reflective language agents under resource constraints. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on theoretical agent regret bounds, not benchmarking methodologies. | |
| GREAT-ESDD: Graph Edge Attention Network for Effectively Steered Delivery Decisions This paper develops a graph edge attention neural solver for dynamic pricing and vehicle routing in same-day delivery operations. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on delivery routing optimization rather than LLM evaluation. | |
| CESNET-TimeSeries26: A Multi-Year Network Traffic Time Series Dataset for Annual Seasonality, Concept Drift, and Long-Term Anomaly Detection This paper presents a multi-year network traffic time-series dataset to evaluate long-term anomaly detection and forecasting models. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates network traffic forecasting models, not language model benchmarks. | |
| YOUDrive: Driving Style as a Composable Direction for End-to-End Autonomous Driving This paper introduces a method to enable composable driving-style control in end-to-end autonomous driving planners using task vectors. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Evaluates driving kinematic trajectories, not language model capabilities. | |
| Depth-progressive 3D seismic velocity model building via 2D generative diffusion models This paper proposes a depth-progressive framework using 2D generative diffusion models to build 3D seismic velocity models efficiently. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Addresses geophysics diffusion modeling, not LLM benchmarking. | |
| Seamless Deployment of Machine Learning Inference at the Far Edge: An Empirical AWS IoT Greengrass Architecture for Industrial Predictive Maintenance This paper evaluates an edge-cloud hybrid architecture using AWS IoT Greengrass for real-time machine learning inference in predictive maintenance. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Focuses on edge IoT architecture rather than language model evaluation. | |
| A Reference Architecture for Integrating Agentic AI into Enterprise Software: Responsibilities, Interactions, and Variability This paper presents a technology-neutral reference architecture for integrating agentic AI capabilities into enterprise software systems. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant System architecture paper, not a benchmark or evaluation methodology. | |
| AI-Native Product Management: A Vision This position paper outlines a conceptual operating model and set of principles for integrating persistent AI agents into product management. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:17 | Not relevant Conceptual framework and vision paper, not an empirical LLM benchmark. |
2,161–2,220 of 3,345← PreviousNext →