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 |
|---|---|---|---|
| SPROUT: Scaling Agentic Reinforcement Learning for Long-Horizon Software Engineering The paper presents SPROUT, a snapshot-based rollout framework to scale reinforcement learning for long-horizon software engineering tasks. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Introduces a training framework rather than an evaluation benchmark. | |
| Scaling Laws for Multimodal Data Mixtures The paper derives multimodal data mixing scaling laws across text, vision, and audio using mixture-of-experts models. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Derives scaling laws for training rather than benchmarking systems. | |
| Query-Adaptive Neuron Steering for LLMs This paper proposes a framework that dynamically steers internal language model neurons based on query-induced activation patterns. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes an inference-time steering mechanism, not an evaluation suite. | |
| Reward-Aligned Distillation This paper presents a bilevel distillation method that weights teacher supervision tokens based on their predicted impact on student reward. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Presents a distillation training method, not an evaluation framework. | |
| Uniform Information Density-Based Preference Optimization This paper proposes a preference optimization algorithm grounded in uniform information density theory to align language model outputs. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes a training optimization method rather than evaluation benchmarks. | |
| VoTSpeech: Decoupling Voice Design from Speech Generation in Instruction-Following TTS This paper presents an instruction-following text-to-speech model that separates voice design from speech generation using an intermediate voice latent. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Presents a TTS model architecture, not evaluation methodology. | |
| Amharic-English Cross-Language Question Answering Using Statistical Machine Translation, Named Entity Recognition, and Semantic Indexing This paper presents a bidirectional Amharic-English cross-language question answering system using machine translation, entity recognition, and semantic indexing. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Presents an information retrieval system, not benchmark methodology. | |
| The Constitution Speaks Nheengatu: An Open MT System and Reproducible Corpus Pipeline for the Amazonian Língua Geral This paper releases an open parallel corpus, speech evaluation set, and translation models for Portuguese and the Amazonian language Nheengatu. Peer-reviewed | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Creates a translation corpus and system, not LLM evaluation. | |
| Training-Free BioNEL-RAG: Advancing Biomedical Entity Linking through Retrieval-Augmented Generation and Large Language Model This paper introduces a training-free retrieval-augmented generation framework for biomedical named entity linking to improve cross-domain adaptability. Peer-reviewed | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes an entity-linking framework, not an evaluation benchmark. | |
| Dependency Reasoning for Aspect Sentiment Quadruple Prediction This paper injects linearized dependency parse trees into language model prompts to improve aspect sentiment quadruple prediction. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Focuses on a task-specific extraction method, not benchmarks. | |
| Agent-Guided Evidence Acquisition for Emotion-Cause Pair Extraction This paper investigates an agent framework that selectively queries expert judgments for borderline emotion-cause pairs to improve extraction accuracy. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Studies extraction strategies rather than LLM benchmarking methods. | |
| Description Complexity Bounds for Analytic Approximations of Lipschitz Functions This paper derives theoretical description complexity bounds for approximating one-dimensional Lipschitz functions using elementary analytic expression trees. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Theoretical math analysis, not LLM benchmarking or evaluation. | |
| FidelityQ: Preserving Predictive Fidelity in Ultra-Low-Bit Language Model Quantization This paper introduces a scalar quantization method using analytic coding to preserve language model predictive fidelity at precisions down to two bits. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes a quantization method rather than benchmark evaluation. | |
| Do Personalized LLMs Need the Whole Conversation? Matching the Full History with Only 5% of Its Text via Relevance-Filtered Context This paper demonstrates that sentence-level relevance retrieval can match full conversation history accuracy in personalized language models using only five percent context. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Evaluates context filtering techniques, not general LLM benchmarks. | |
| HiPAD: A Unified Gait, Audio-Visual, and Personality Fusion Network for Depression Detection, with a Hierarchical Personality Ablation and the Calibration Limits of Domain-Adapted LLMs This paper presents a multimodal fusion network for depression detection combining gait, audio-visual data, and textual personality descriptions. Peer-reviewed | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Presents a challenge submission network rather than evaluation methodology. | |
| HodgePhaseNet: Discrete Hodge Decomposition of Semantic Flows in Transformer Attention This paper introduces a discrete Hodge decomposition method to separate semantic flow transitions within transformer attention mechanisms. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Presents an attention architecture, not benchmark evaluation methodology. | |
| What Drives AI Citations? Content, Retrieval and Model Knowledge in Generative Engine Optimization (GEO) This paper examines how content matching, retrieval mechanics, and model knowledge affect web page citations in generative search engines. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Empirical analysis of search engines rather than evaluation methodology. | |
| PAT: A Personality-Driven Augmentation and Transfer Learning Framework for Depression Detection on Social Media This paper presents a personality-driven augmentation and contrastive learning framework to improve depression detection from social media posts. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes a detection framework rather than a general evaluation benchmark. | |
| Compute-Optimal Scaling Laws for Spiking Language Models The paper presents an empirical scaling law study measuring parameter and compute trade-offs in spiking neural network language models. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Analyzes compute scaling laws rather than LLM evaluation benchmarks. | |
| Where Did We Go Wrong? Localizing Errors to Repair Multi-Turn Conversations The paper presents a framework to probabilistically localize user-intent misinterpretations and repair conversational states in multi-turn assistant systems. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Presents an error-recovery framework rather than an evaluation benchmark. | |
| Learning Proposition-Level Links from Sentence-Level Supervision for Neuro-Symbolic NLI The paper proposes a trainable proposition linker that learns atom-level semantic links from sentence-level supervision for neuro-symbolic inference. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes a linking architecture rather than an evaluation benchmark. | |
| Active Learning for Finding a Needle in a Haystack The paper evaluates whether active learning strategies can overcome initial selection biases when identifying rare document classes in large text corpora. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Evaluates active learning algorithms rather than LLM benchmark design. | |
| Word-Aligned Prosody Reflects Speaker Differences More Than Depression or PTSD Severity The paper investigates whether word-aligned prosodic speech measurements reflect speaker identity traits more strongly than depression or PTSD severity. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Evaluates acoustic biomarkers rather than language model benchmarks. | |
| SAFE-KV: Causal Suffix Repair for Behavior-Preserving Non-Prefix KV Cache Reuse in RAG The paper presents a repair method that re-encodes causal suffixes to enable safe non-prefix KV-cache reuse in retrieval-augmented generation. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Introduces a cache repair technique rather than an evaluation benchmark. | |
| Do LLMs differentiate linguistic traits internally?: A layer-wise analysis The paper conducts layer-wise representational analysis to determine how language models internally distinguish conventional meanings from idiomatic traits. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Examines internal interpretability probing rather than LLM benchmarking. | |
| From Candidate Tokens to Final Outcomes: Decoding Relative Long-Horizon Candidate Value from Reasoning Model Representations The paper constructs a dataset to show that relative candidate token value in multi-step reasoning is linearly decodable from model representations. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Investigates internal representation probing rather than evaluation benchmark integrity. | |
| AlphaPilot: Multi-Turn Tool-Using LLM Agents for Adaptive Alpha Mining with Reinforcement Learning The paper introduces a reinforcement learning framework that trains tool-using agents to generate and refine financial alpha factor formulas. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes an alpha-mining agent rather than an LLM benchmark. | |
| Concept Representations in Vector Space: Quantifying Sharedness and Idiosyncrasy of Meaning The paper develops a measurement framework using sentence embeddings to quantify shared versus individual-specific variation in concept definitions. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Focuses on psycholinguistic measurement rather than LLM benchmarking. | |
| A Step in the Right Place: Understanding and Optimizing Timestep Placement in Continuous Diffusion Language Models The paper introduces an algorithm to optimize denoising timestep placement in continuous diffusion language models to improve generation quality. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Focuses on sampling algorithms rather than evaluation benchmarks. | |
| Apertus 1.5: From Open Text Models to Open Multimodal Reasoners The paper presents open-source multimodal models extended to process text, images, and audio with long-context reasoning capabilities. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Evaluates a new model release rather than benchmarking methodology. | |
| An Interpretable Linguistically-Grounded Analysis of Scene Saliency in Movie Screenplays The paper evaluates whether uniform information density and specific linguistic features predict salient scenes in movie screenplays. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Studies linguistic salience rather than LLM benchmarking methodologies. | |
| The Fragility and Dynamics of LLM Pruning-Recovery Steps The paper analyzes how pruning and subsequent recovery stages interact, finding instruction-following collapses earlier than domain knowledge during model compression. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Evaluates pruning recipes rather than creating evaluation benchmarks. | |
| Towards Robust Calibration Dataset for Domain-General Mixture of Experts Pruning The paper studies how calibration dataset size, language, and data distribution affect post-training expert pruning in mixture-of-experts models. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Studies pruning calibration data rather than evaluation benchmark design. | |
| The Gradient That Top-k Distillation Never Computes The paper analyzes gradient dynamics in top-k distillation, showing that renormalization prevents optimization along directions controlling tail token probabilities. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Analyzes training distillation gradients rather than evaluation benchmarks. | |
| The Mathematical Collective Unconscious: Jungian Archetypes within the Latent Space The paper examines whether large language model embedding spaces naturally form semantic clusters resembling Jungian archetypes using synthetic and human-authored texts. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Latent space exploratory study, not a benchmark. | |
| Towards Autopilot Labeling for Short Video Content Governance The paper presents a multi-stage automated labeling framework using multimodal models and agents to annotate short videos for content moderation. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Presents an automated labeling system, not an evaluation benchmark. | |
| Preserving Updates in Low-Bit Recurrent State Quantization This paper proposes a low-rank bypass to accumulate small updates and reduce compounding errors in quantized recurrent model states. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Does not propose benchmarks or evaluation methodologies. | |
| TL;QR: Querying Temporal Discursive Patterns in Textual Corpora TL;QR pairs language-model passage judgments with a symbolic temporal logic engine to track evolving patterns in textual corpora. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes a querying engine, not a model benchmark. | |
| Enhancing Zero-Shot Domain Transfer in Low-Resource Sentiment Analysis: A Case Study in Marathi This paper studies multi-domain training setups to improve zero-shot sentiment classification transfer across unseen domains in Marathi. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Empirical study on domain transfer, not benchmark design. | |
| Robust Marathi Sentiment Analysis using Domain-Adaptive Pre-training and Data Augmentation Approaches The authors evaluate domain-adaptive pretraining and back-translation data augmentation to improve sentiment classification in Marathi. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Applies standard training techniques, no novel evaluation paradigm. | |
| RSIPO: Recursive Self-Improving Policy Optimization for Agentic Reinforcement Learning via Skill-Policy Co-Evolution RSIPO introduces a reinforcement learning framework where agentic skills and policy models co-evolve using self-extracted trajectory experience. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Does not introduce a new benchmark or judge method. | |
| Heterogeneous Structured Pruning Under Resource Constraints for Hardware-Efficient Large Language Models Heterogeneous Structured Pruning selects sublayer-specific sparsity patterns guided by latency and memory models for hardware-efficient inference. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Does not introduce LLM benchmarks or evaluation methods. | |
| Latent Taxonomies: How Property Induction Develops in Language Model Pre-training This study analyzes how category-based property induction emerges in language models during pretraining using a synthetic hierarchical taxonomy. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Studies pre-training phenomena rather than testing benchmarks. | |
| KhyontekAI at CheckThat! 2026: Multilingual Dense Retrieval and Cross-Encoder Reranking for Scientific Web Claims The authors present a multilingual dense retrieval and reranking framework to link scientific web claims to their source publications. Peer-reviewed | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Presents a competition system, not an evaluation framework. | |
| Binding Phoneme Identity to Position in Neural Models and the Human Brain The study tests whether neural models bind phoneme identity to serial position using state direction and magnitude representations. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Does not introduce benchmarks or LLM evaluation frameworks. | |
| RIVO: Lightweight Adaptive Rerouting for Deep Research Agents RIVO is a lightweight plug-in framework that adaptively reroutes research agent trajectories to prevent premature stopping or wasted steps. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Does not introduce a new benchmark or evaluation method. | |
| More Reasoning Is Not Always Better: Executing Clinical Guidelines with LLMs and Rules MedDEx combines language models for patient data extraction with deterministic rule engines to execute clinical treatment guidelines accurately and efficiently. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes a hybrid decision framework rather than evaluation benchmarks. | |
| Natural Language Processing for Aligning Readings with Learning Objectives and Assessing Topic Balance in Dental Education The paper develops a natural language processing workflow to check reading list alignment with curriculum objectives and topic balance. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Presents a domain-specific curriculum workflow, not an LLM benchmark. | |
| Scaling Repeated-Data Pretraining: Batch Size, Regularization, and GPU-Time Tradeoffs This paper investigates batch size, regularization methods, and runtime tradeoffs when pretraining language models on repeated finite data. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Investigates pretraining optimization rather than benchmark development. | |
| Mixed Feelings, Single Polarity The authors examine failure points when projecting complex emotion dataset labels onto single positive or negative valence polarities. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Examines classical sentiment dataset annotations rather than LLM benchmarks. | |
| PAIR: Memory-Efficient and Fast Inference with Reading--Reasoning Fusion PAIR splits context processing and answer reasoning between two different-sized models, fusing predictions at the logit level for efficient inference. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Presents an inference architecture rather than evaluation benchmarks. | |
| Scaling Multi-Agent Intelligence with Substrate-Centric Coordination ScaleMAS coordinates multi-agent systems using an asynchronous shared datastore with limited search to improve scaling and fault tolerance. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes a multi-agent framework rather than evaluation benchmarks. | |
| Hashrope: A Persistent Data Structure for Mechanisms of LLM Context Management This paper presents Hashrope, a unified persistent data structure supporting incremental edits, branching, and cache reuse for language model serving. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Introduces a context management data structure, not an evaluation benchmark. | |
| Think Before You Speak: Enabling Training-Free Latent Communication Between Language Models This paper presents CacheFuse, a training-free technique enabling heterogeneous language models to communicate directly by sharing key-value cache states. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes a collaboration method rather than benchmark evaluation methodologies. | |
| DeepThink: A Conceptual Model for Knowledge-Addressable Recurrent Language Modeling This paper proposes DeepThink, a conceptual architecture that decouples knowledge capacity, activation width, and recurrent depth in language models. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Introduces a conceptual model architecture, not evaluation benchmarks. | |
| Error Correction in Trained TTT Layers This paper analyzes prediction-derived error updates and test-time gradient objectives in trained test-time-training layers. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Studies gradient mechanics in TTT layers rather than evaluation benchmarks. | |
| Deep Speech Representations Track Word Timing along a Low-Dimensional Circular Trajectory This paper demonstrates that deep speech representations track word timing along low-dimensional circular trajectories across multiple model architectures. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Analyzes representation geometry rather than benchmark evaluation frameworks. | |
| ATTCK-R: Knowledge-Aware Long-Horizon Jailbreaking for Protected LLMs This paper presents ATTCK-R, a planning-driven multi-turn red teaming framework to evaluate language models guarded by prompt safety defenses. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes an adversarial red-teaming method, not general evaluation benchmarks. | |
| RIFT: Token Influence Follows the Path of Least Resistance This paper proposes RIFT, a method using electrical resistance analogies to trace token influence through transformer attention graphs. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Introduces an interpretability tool rather than a model evaluation benchmark. | |
| A Production-Grade, High-Efficiency Agentic RL Framework for Frontier LLMs with Validated Recipes This paper introduces LoongSage, an asynchronous reinforcement learning framework designed to run multi-turn agent tasks with reduced system overhead. Preprint | LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18 | Not relevant Proposes an RL training framework rather than evaluation benchmarks. |
1,981–2,040 of 3,345← PreviousNext →