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
Inject a Turn, Not a Token: Learning Where to Redirect LLM Reasoning for Self-Correction Introduces UPTURN, a framework that trains a small classifier to inject redirection prompts as new user turns during reasoning failures. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Evaluates general reasoning on math and science benchmarks rather than coding or software repair.
Inject a Turn, Not a Token: Learning Where to Redirect LLM Reasoning for Self-Correction Introduces UPTURN, a framework that trains a small classifier to inject redirection prompts as new user turns during reasoning failures. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Investigates test-time reasoning self-correction without using external document or database retrieval.
Not All Dimensions Are Equal: Outlier-Aware Attention Linearization with Grouped Quadratic Kernels Presents Outlier-Aware Attention Linearization, using quadratic approximations to retain outlier dimensions during efficient Transformer attention evaluation. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses inference efficiency rather than deployed AI safety, guardrails, or alignment.
Not All Dimensions Are Equal: Outlier-Aware Attention Linearization with Grouped Quadratic Kernels Presents Outlier-Aware Attention Linearization, using quadratic approximations to retain outlier dimensions during efficient Transformer attention evaluation. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Develops a linearized attention mechanism rather than evaluation benchmarks or LLM-as-a-judge protocols.
Not All Dimensions Are Equal: Outlier-Aware Attention Linearization with Grouped Quadratic Kernels Presents Outlier-Aware Attention Linearization, using quadratic approximations to retain outlier dimensions during efficient Transformer attention evaluation. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on attention kernel approximation rather than autonomous agent planning or actions.
Variance-Reduced Policy Gradient is Nearly Minimax-Optimal Introduces Loopless Variance-Reduced Policy Gradient, proving near-optimal sample complexity under weak gradient domination using first-order updates. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Theoretical analysis of policy gradient sample complexity, not an LLM benchmark or evaluation study.
Variance-Reduced Policy Gradient is Nearly Minimax-Optimal Introduces Loopless Variance-Reduced Policy Gradient, proving near-optimal sample complexity under weak gradient domination using first-order updates. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on generic RL theoretical convergence rates rather than AI safety or alignment risks.
Variance-Reduced Policy Gradient is Nearly Minimax-Optimal Introduces Loopless Variance-Reduced Policy Gradient, proving near-optimal sample complexity under weak gradient domination using first-order updates. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Presents core reinforcement learning theory rather than LLM planning, multi-agent systems, or tool use.
Variance-Reduced Policy Gradient is Nearly Minimax-Optimal Introduces Loopless Variance-Reduced Policy Gradient, proving near-optimal sample complexity under weak gradient domination using first-order updates. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Analyzes policy gradient theoretical bounds rather than efficient model inference, serving, or compression.
Calibration in One Go: Jointly Learning Accurate and Calibrated Neural Networks Employs a bilevel Stackelberg formulation to jointly optimize classification accuracy and confidence calibration without separate calibration datasets. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Focuses on prediction calibration rather than model compression, quantization, or inference acceleration.
Calibration in One Go: Jointly Learning Accurate and Calibrated Neural Networks Employs a bilevel Stackelberg formulation to jointly optimize classification accuracy and confidence calibration without separate calibration datasets. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Addresses model training loss formulations rather than LLM benchmarking or evaluation frameworks.
Calibration in One Go: Jointly Learning Accurate and Calibrated Neural Networks Employs a bilevel Stackelberg formulation to jointly optimize classification accuracy and confidence calibration without separate calibration datasets. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Investigates neural network output calibration, not multi-step LLM agents or tool usage.
A Baseline Assessment for the Agentic Web Technology Stack Benchmarks combinations of language models, communication protocols, and agent frameworks on web interaction tasks. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Evaluates web-navigation agents on Mind2Web rather than models writing, reviewing, or fixing software.
A Baseline Assessment for the Agentic Web Technology Stack Benchmarks combinations of language models, communication protocols, and agent frameworks on web interaction tasks. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on system stack evaluation rather than safety vulnerabilities, guardrails, or jailbreaks.
Adaptive DHiSparse: Speculative Decoding over a Load-Adaptive KV Hierarchy Combines load-adaptive hierarchical key-value caching with speculative decoding to accelerate long-context language model serving. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses system performance and throughput rather than safety, alignment, or guardrails.
Adaptive DHiSparse: Speculative Decoding over a Load-Adaptive KV Hierarchy Combines load-adaptive hierarchical key-value caching with speculative decoding to accelerate long-context language model serving. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Investigates inference serving throughput rather than benchmark methodology or LLM evaluation frameworks.
Adaptive DHiSparse: Speculative Decoding over a Load-Adaptive KV Hierarchy Combines load-adaptive hierarchical key-value caching with speculative decoding to accelerate long-context language model serving. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on inference hardware optimization and KV cache serving rather than autonomous agent systems.
Adaptive DHiSparse: Speculative Decoding over a Load-Adaptive KV Hierarchy Combines load-adaptive hierarchical key-value caching with speculative decoding to accelerate long-context language model serving. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Evaluates text LLMs on context serving without multimodal document, vision, or audio components.
Adaptive DHiSparse: Speculative Decoding over a Load-Adaptive KV Hierarchy Combines load-adaptive hierarchical key-value caching with speculative decoding to accelerate long-context language model serving. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Deals with internal KV cache hierarchy retrieval during inference, not external knowledge RAG.
Adaptive DHiSparse: Speculative Decoding over a Load-Adaptive KV Hierarchy Combines load-adaptive hierarchical key-value caching with speculative decoding to accelerate long-context language model serving. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Optimizes serving infrastructure rather than software engineering or code intelligence.
Amortized Counterfactual Prediction with Graph Neural Networks Presents a graph neural process framework to answer counterfactual queries directly from observational data without prior causal graph discovery. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Proposes a prediction architecture rather than an LLM evaluation framework or benchmark.
Amortized Counterfactual Prediction with Graph Neural Networks Presents a graph neural process framework to answer counterfactual queries directly from observational data without prior causal graph discovery. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Presents general causal inference without focusing on deployed AI safety, guardrails, or alignment.
Amortized Counterfactual Prediction with Graph Neural Networks Presents a graph neural process framework to answer counterfactual queries directly from observational data without prior causal graph discovery. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on causal inference via GNNs, not multi-step LLM planning or tool use.
Amortized Counterfactual Prediction with Graph Neural Networks Presents a graph neural process framework to answer counterfactual queries directly from observational data without prior causal graph discovery. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Does not address model compression, quantization, distillation, or efficient inference serving.
LAPT-Scale: Scaling Embodied Policies with Latent Action Pre-Training on Human Videos Pre-trains embodied robot policies on human egocentric videos using learned continuous latent action representations. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Does not study safety, guardrails, jailbreaks, or deployment reliability risks.
LAPT-Scale: Scaling Embodied Policies with Latent Action Pre-Training on Human Videos Pre-trains embodied robot policies on human egocentric videos using learned continuous latent action representations. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Proposes a training method for robotics rather than a benchmark or evaluation methodology.
LAPT-Scale: Scaling Embodied Policies with Latent Action Pre-Training on Human Videos Pre-trains embodied robot policies on human egocentric videos using learned continuous latent action representations. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:28	Not relevant Focuses on embodied robotic manipulation rather than industrial inspection, defect detection, or factory visual tasks.
Orthros: Phase-Aware Heterogeneous Attention for Efficient Transformers Proposes an attention mechanism that switches from linear complexity during prefilling to full attention during token decoding. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Proposes an attention mechanism rather than multi-step tool-using language model agents.
Orthros: Phase-Aware Heterogeneous Attention for Efficient Transformers Proposes an attention mechanism that switches from linear complexity during prefilling to full attention during token decoding. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Focuses on architectural design for attention rather than LLM evaluation benchmarks.
Orthros: Phase-Aware Heterogeneous Attention for Efficient Transformers Proposes an attention mechanism that switches from linear complexity during prefilling to full attention during token decoding. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Optimizes computational complexity of attention rather than safety or reliability guardrails.
SafeDiSTL: Decomposing Joint Constraints for Safe Multi-Agent Reinforcement Learning Decomposes joint temporal logic constraint violations to allocate credit among agents in cooperative reinforcement learning. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Proposes an RL credit assignment method rather than an LLM evaluation benchmark.
SafeDiSTL: Decomposing Joint Constraints for Safe Multi-Agent Reinforcement Learning Decomposes joint temporal logic constraint violations to allocate credit among agents in cooperative reinforcement learning. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Focuses on constrained reinforcement learning rather than small or quantized models.
A Privileged Teacher That Only Scores Cannot Search: Critique-Conditioned On-Policy Self-Distillation for Reasoning Examines error localization and credit assignment failures in critique-conditioned on-policy self-distillation for mathematical reasoning. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Deals with general mathematical/reasoning distillation rather than software engineering or code LLMs.
A Privileged Teacher That Only Scores Cannot Search: Critique-Conditioned On-Policy Self-Distillation for Reasoning Examines error localization and credit assignment failures in critique-conditioned on-policy self-distillation for mathematical reasoning. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Focuses strictly on language-based reasoning traces without multimodal inputs.
A Privileged Teacher That Only Scores Cannot Search: Critique-Conditioned On-Policy Self-Distillation for Reasoning Examines error localization and credit assignment failures in critique-conditioned on-policy self-distillation for mathematical reasoning. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on training reasoning models via distillation rather than autonomous multi-step agents.
A Privileged Teacher That Only Scores Cannot Search: Critique-Conditioned On-Policy Self-Distillation for Reasoning Examines error localization and credit assignment failures in critique-conditioned on-policy self-distillation for mathematical reasoning. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Studies self-distillation for reasoning rather than retrieval-augmented generation systems.
A Privileged Teacher That Only Scores Cannot Search: Critique-Conditioned On-Policy Self-Distillation for Reasoning Examines error localization and credit assignment failures in critique-conditioned on-policy self-distillation for mathematical reasoning. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Investigates distillation credit signals rather than LLM benchmarking or evaluation frameworks.
Matched-Context Group Policy Optimization for LLM Agent Training Constructs context-matched preference comparisons from sibling rollouts to improve credit assignment during language model agent training. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Introduces a reinforcement learning training method rather than an evaluation benchmark.
Matched-Context Group Policy Optimization for LLM Agent Training Constructs context-matched preference comparisons from sibling rollouts to improve credit assignment during language model agent training. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on task performance training rather than AI safety, robustness, or guardrails.
Matched-Context Group Policy Optimization for LLM Agent Training Constructs context-matched preference comparisons from sibling rollouts to improve credit assignment during language model agent training. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Focuses on agent policy optimization rather than model quantization or inference efficiency.
When Edge Independence Fails: Joint Graph Diffusion with Latent Sociability Priors Introduces a latent sociability prior to mitigate edge-independence issues in discrete graph diffusion models. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on graph diffusion theory rather than LLM-based autonomous agents.
When Edge Independence Fails: Joint Graph Diffusion with Latent Sociability Priors Introduces a latent sociability prior to mitigate edge-independence issues in discrete graph diffusion models. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Studies graph generation priors rather than language model evaluation frameworks.
When Edge Independence Fails: Joint Graph Diffusion with Latent Sociability Priors Introduces a latent sociability prior to mitigate edge-independence issues in discrete graph diffusion models. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses graph generative modeling theory rather than AI safety, alignment, or guardrails.
When Edge Independence Fails: Joint Graph Diffusion with Latent Sociability Priors Introduces a latent sociability prior to mitigate edge-independence issues in discrete graph diffusion models. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Deals with graph generation quality rather than model efficiency or compression.
DualQuant: Dual-scale Calibration-free Weight-and-Activation Quantization for LLMs Proposes a calibration-free post-training quantization method that jointly optimizes weight and activation scaling for 4-bit model inference. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses model compression and hardware speedups rather than safety or guardrails.
DualQuant: Dual-scale Calibration-free Weight-and-Activation Quantization for LLMs Proposes a calibration-free post-training quantization method that jointly optimizes weight and activation scaling for 4-bit model inference. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on low-bit model quantization rather than multi-step agent systems.
DualQuant: Dual-scale Calibration-free Weight-and-Activation Quantization for LLMs Proposes a calibration-free post-training quantization method that jointly optimizes weight and activation scaling for 4-bit model inference. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Develops a quantization method rather than an evaluation benchmark for language models.
ReplaySSM: Cache SSM Inputs, Not State Replaces frequent state memory updates in state-space models by caching inputs and periodically checkpointing state during decoding. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on runtime throughput optimization rather than AI safety or alignment.
ReplaySSM: Cache SSM Inputs, Not State Replaces frequent state memory updates in state-space models by caching inputs and periodically checkpointing state during decoding. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Optimizes state-space model inference rather than autonomous agent planning or actions.
ReplaySSM: Cache SSM Inputs, Not State Replaces frequent state memory updates in state-space models by caching inputs and periodically checkpointing state during decoding. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Proposes an inference acceleration mechanism rather than benchmark or evaluation methodologies.
UkrMeme: Multimodal Explainable Persuasion Detection in Ukrainian Memes Introduces a Ukrainian meme dataset and evaluates vision-language models on detecting persuasive and propaganda techniques. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Focuses on political memes, completely unrelated to code generation or software tasks.
UkrMeme: Multimodal Explainable Persuasion Detection in Ukrainian Memes Introduces a Ukrainian meme dataset and evaluates vision-language models on detecting persuasive and propaganda techniques. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Evaluates zero-shot VLM classification on memes rather than agentic planning or tools.
UkrMeme: Multimodal Explainable Persuasion Detection in Ukrainian Memes Introduces a Ukrainian meme dataset and evaluates vision-language models on detecting persuasive and propaganda techniques. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:28	Not relevant Analyzes online multimodal memes rather than industrial vision inspection tasks.
UkrMeme: Multimodal Explainable Persuasion Detection in Ukrainian Memes Introduces a Ukrainian meme dataset and evaluates vision-language models on detecting persuasive and propaganda techniques. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Does not address small models, model compression, or inference optimization.
UkrMeme: Multimodal Explainable Persuasion Detection in Ukrainian Memes Introduces a Ukrainian meme dataset and evaluates vision-language models on detecting persuasive and propaganda techniques. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Studies meme persuasion detection without retrieval-augmented generation mechanisms.
Your Discrete Graph Diffusion Model is Secretly Memorizing Demonstrates that discrete graph diffusion models exhibit conditional memorization when reverse trajectories are conditioned on corrupted training graphs. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Investigates memorization in graph diffusion rather than LLM evaluation or benchmarks.
Your Discrete Graph Diffusion Model is Secretly Memorizing Demonstrates that discrete graph diffusion models exhibit conditional memorization when reverse trajectories are conditioned on corrupted training graphs. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Studies graph diffusion models rather than LLM-based autonomous agent architectures.
Your Discrete Graph Diffusion Model is Secretly Memorizing Demonstrates that discrete graph diffusion models exhibit conditional memorization when reverse trajectories are conditioned on corrupted training graphs. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Deals with graph generation theory, not model compression or efficient serving.
FORTE: A Production Framework for Feature Evaluation and Deprecation in Industrial Recommender Systems Describes an industrial framework for evaluating, ablating, and deprecating features in production recommender systems without performance regressions. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Presents recommender system infrastructure rather than language model planning agents.
FORTE: A Production Framework for Feature Evaluation and Deprecation in Industrial Recommender Systems Describes an industrial framework for evaluating, ablating, and deprecating features in production recommender systems without performance regressions. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Focuses on feature lifecycle in recommender pipelines, not model compression or small models.

17,221–17,280 of 17,313← Previous Next →