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
Optimal random quantisers for spherically symmetric distributions Analyzes optimal random quantisers for high-dimensional spherically symmetric distributions to reduce distortion with moderate sample sizes. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Theoretical statistics paper, not related to AI safety, security, guardrails, or alignment.
In-Ride Alcohol-Impairment Detection in E-Scooterists with False-Alarm Control Develops an onboard inertial and throttle sensor detector with false-alarm controls to identify alcohol-impaired e-scooter riders during trips. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Presents a sensor-based classifier for e-scooters, not a benchmark or framework for LLM evaluation.
In-Ride Alcohol-Impairment Detection in E-Scooterists with False-Alarm Control Develops an onboard inertial and throttle sensor detector with false-alarm controls to identify alcohol-impaired e-scooter riders during trips. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on edge sensor classification for intoxication, not language model compression or efficiency.
In-Ride Alcohol-Impairment Detection in E-Scooterists with False-Alarm Control Develops an onboard inertial and throttle sensor detector with false-alarm controls to identify alcohol-impaired e-scooter riders during trips. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Deals with physical vehicle safety via sensor anomaly detection, not AI model deployment safety or alignment.
Compile the Table: Query-Calibrated Operator Compression for Tabular In-Context Learning Compresses tabular in-context learning key-value caches into query-calibrated prototypes to improve throughput without sacrificing accuracy. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Proposes a KV cache compression algorithm for tabular learning, not an LLM evaluation benchmark.
Compile the Table: Query-Calibrated Operator Compression for Tabular In-Context Learning Compresses tabular in-context learning key-value caches into query-calibrated prototypes to improve throughput without sacrificing accuracy. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses latency and cache compression efficiency, not system safety, alignment, or hallucination guardrails.
Softmax Attention on Gaussian Mixtures: Linear When It Can, Selective When It Must Analyzes the infinite-prompt limit of softmax attention on Gaussian mixtures to explain when attention behaves linearly or selectively. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Theoretical analysis of attention mechanisms on Gaussian mixtures, not LLM benchmark evaluation or testing.
Softmax Attention on Gaussian Mixtures: Linear When It Can, Selective When It Must Analyzes the infinite-prompt limit of softmax attention on Gaussian mixtures to explain when attention behaves linearly or selectively. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Mathematical analysis of attention limits, not operational AI safety, guardrails, or deployment reliability.
Softmax Attention on Gaussian Mixtures: Linear When It Can, Selective When It Must Analyzes the infinite-prompt limit of softmax attention on Gaussian mixtures to explain when attention behaves linearly or selectively. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Theoretical study of softmax attention properties, not practical small model design, quantization, or distillation.
PRAXIS: Learning Dynamics of Self-Improving Models with Symbolic Archives Models self-improving systems with symbolic archives as dynamical processes, establishing theoretical optimization bounds and empirical stability. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on co-evolutionary learning theory and symbolic archives, not model compression or efficient inference.
PRAXIS: Learning Dynamics of Self-Improving Models with Symbolic Archives Models self-improving systems with symbolic archives as dynamical processes, establishing theoretical optimization bounds and empirical stability. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Theoretical formulation of nonstationary objective drift, not AI alignment, jailbreaks, hallucination, or deployment safety.
PRAXIS: Learning Dynamics of Self-Improving Models with Symbolic Archives Models self-improving systems with symbolic archives as dynamical processes, establishing theoretical optimization bounds and empirical stability. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Studies optimization dynamics in self-improving systems, not general LLM evaluation frameworks or benchmark integrity.
Self-Supervised Speech Representations for Cross-Speaker Dysarthria Detection During Awake Craniotomy Evaluates a pipeline combining speaker diarization and self-supervised speech representations to detect dysarthria during awake craniotomy procedures. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Evaluates a specialized acoustic clinical pipeline, not general LLM evaluation benchmarks or evaluation methodologies.
Self-Supervised Speech Representations for Cross-Speaker Dysarthria Detection During Awake Craniotomy Evaluates a pipeline combining speaker diarization and self-supervised speech representations to detect dysarthria during awake craniotomy procedures. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on clinical speech classification representations, not model compression or efficient model architectures.
Self-Supervised Speech Representations for Cross-Speaker Dysarthria Detection During Awake Craniotomy Evaluates a pipeline combining speaker diarization and self-supervised speech representations to detect dysarthria during awake craniotomy procedures. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Pertains to intra-operative medical diagnostics rather than deployment guardrails, hallucination, or AI safety risks.
Recovery Guarantees for Posterior Sampling of One-Bit Compressed Sensing Establishes theoretical sample complexity bounds for posterior sampling in noisy one-bit compressed sensing under prior distribution mismatch. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on mathematical signal reconstruction guarantees, not AI deployment safety, robustness, or alignment risks.
Recovery Guarantees for Posterior Sampling of One-Bit Compressed Sensing Establishes theoretical sample complexity bounds for posterior sampling in noisy one-bit compressed sensing under prior distribution mismatch. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Addresses signal processing recovery bounds in compressed sensing, not model quantization, distillation, or efficient inference.
Recovery Guarantees for Posterior Sampling of One-Bit Compressed Sensing Establishes theoretical sample complexity bounds for posterior sampling in noisy one-bit compressed sensing under prior distribution mismatch. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Theoretical study on sample complexity in compressed sensing, not LLM evaluation or benchmark design.
Structural and Representational Agency in LLM-Driven Timbre Generation and Interpolation Examines the disparity between reachability and self-reported parameters in an LLM-driven audio interface for timbre generation. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Analyzes user interfaces and synthesizer parameter accuracy in audio generation, not general LLM benchmarking methods.
Structural and Representational Agency in LLM-Driven Timbre Generation and Interpolation Examines the disparity between reachability and self-reported parameters in an LLM-driven audio interface for timbre generation. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Does not explore small models, model distillation, quantization, or efficient deployment techniques.
Structural and Representational Agency in LLM-Driven Timbre Generation and Interpolation Examines the disparity between reachability and self-reported parameters in an LLM-driven audio interface for timbre generation. Preprint	Multimodal models via arxiv · 12 Oct, 16:29	Not relevant Examines audio synthesis interface parameters and generative timbre, not multimodal foundation models or vision-language reasoning.
Structural and Representational Agency in LLM-Driven Timbre Generation and Interpolation Examines the disparity between reachability and self-reported parameters in an LLM-driven audio interface for timbre generation. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Covers creative audio interface discrepancies rather than AI safety, red-teaming, guardrails, or deployment reliability.
DADP: Dynamic Activity-Dependent Pruning, A Reverse Hebbian-Inspired Structural Pruning Method Proposes a dynamic structural pruning method using biologically inspired plasticity to reduce neural network parameters during training. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Proposes a neural network pruning technique rather than an LLM evaluation benchmark or evaluation framework.
DADP: Dynamic Activity-Dependent Pruning, A Reverse Hebbian-Inspired Structural Pruning Method Proposes a dynamic structural pruning method using biologically inspired plasticity to reduce neural network parameters during training. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on model sparsity and parameter efficiency, not safety, security, jailbreak mitigation, or alignment.
Conditional Kernel Stein Discrepancy Develops a kernel Stein discrepancy framework using operator-valued kernels to assess conditional goodness-of-fit for unnormalized models. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Focuses on statistical hypothesis testing, not AI deployment safety, hallucination mitigation, or robustness guardrails.
Conditional Kernel Stein Discrepancy Develops a kernel Stein discrepancy framework using operator-valued kernels to assess conditional goodness-of-fit for unnormalized models. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Theoretical statistical testing paper without any connection to model compression, small models, or efficient inference.
Conditional Kernel Stein Discrepancy Develops a kernel Stein discrepancy framework using operator-valued kernels to assess conditional goodness-of-fit for unnormalized models. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Presents statistical goodness-of-fit testing theory using kernel Stein discrepancies, not LLM benchmarking or evaluation methodology.
Understanding Latent-Dimension Scaling in Dynamical-System Learning through Spectral Reliability Analyzes how latent-dimension scaling impacts dynamical-system learning and rollout error using Koopman operator spectral reliability metrics. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on theoretical dynamical systems and Koopman operator spectral analysis, not LLM evaluation benchmarks.
Understanding Latent-Dimension Scaling in Dynamical-System Learning through Spectral Reliability Analyzes how latent-dimension scaling impacts dynamical-system learning and rollout error using Koopman operator spectral reliability metrics. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Deals with dynamical-system state representation scaling, not language model compression, quantization, or efficient serving.
Understanding Latent-Dimension Scaling in Dynamical-System Learning through Spectral Reliability Analyzes how latent-dimension scaling impacts dynamical-system learning and rollout error using Koopman operator spectral reliability metrics. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses spectral reliability in physics/dynamical systems, not deployment safety, guardrails, or hallucination in AI systems.
Learning structured linear dynamical systems from missing observations Establishes non-asymptotic statistical error bounds and optimization convergence for learning structured linear dynamical systems from highly subsampled observations. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Theoretical recovery of dynamical systems, not small language models.
Learning structured linear dynamical systems from missing observations Establishes non-asymptotic statistical error bounds and optimization convergence for learning structured linear dynamical systems from highly subsampled observations. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on linear dynamical systems with missing observations, not LLM evaluation.
Learning structured linear dynamical systems from missing observations Establishes non-asymptotic statistical error bounds and optimization convergence for learning structured linear dynamical systems from highly subsampled observations. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Statistical dynamical system recovery, not deployment safety or alignment.
Automated Assembly Instruction Generation from CAD Models Using Grounded Large Language Models: A Human-in-the-Loop Framework Builds a human-in-the-loop framework using grounded language models to automatically generate natural-language assembly instructions from CAD models. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Presents an instruction generation framework, not an LLM benchmark.
Automated Assembly Instruction Generation from CAD Models Using Grounded Large Language Models: A Human-in-the-Loop Framework Builds a human-in-the-loop framework using grounded language models to automatically generate natural-language assembly instructions from CAD models. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Applies standard LLMs to CAD models, not model efficiency or distillation.
RobustLDS: Learning linear dynamical systems under adversarial corruptions Develops robust estimators and optimization algorithms to learn linear dynamical systems from trajectories corrupted by adversarial outliers. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Addresses identification of dynamical systems, not LLM evaluation.
RobustLDS: Learning linear dynamical systems under adversarial corruptions Develops robust estimators and optimization algorithms to learn linear dynamical systems from trajectories corrupted by adversarial outliers. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Theoretical estimation of dynamical systems, not small model efficiency.
Cost-Aware Mixture-of-Experts Coordination for Model Markets Proposes a market-level mixture-of-experts coordination framework using cost-aware gating and revenue allocation across heterogeneous model services. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Economic coordination mechanism, not deployment safety or jailbreak prevention.
Cost-Aware Mixture-of-Experts Coordination for Model Markets Proposes a market-level mixture-of-experts coordination framework using cost-aware gating and revenue allocation across heterogeneous model services. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Market coordination for mixture-of-experts, not LLM benchmarks.
Puffin: Probabilistic Learning of Spatial Detail From Coarse Observations Presents a probabilistic framework to disaggregate coarse socioeconomic observations into fine spatial scales using satellite embeddings. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Deals with geospatial count estimation, not AI deployment safety.
Puffin: Probabilistic Learning of Spatial Detail From Coarse Observations Presents a probabilistic framework to disaggregate coarse socioeconomic observations into fine spatial scales using satellite embeddings. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on spatial statistical disaggregation, not LLM benchmarking.
Puffin: Probabilistic Learning of Spatial Detail From Coarse Observations Presents a probabilistic framework to disaggregate coarse socioeconomic observations into fine spatial scales using satellite embeddings. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Socioeconomic spatial modeling, not efficient language models or pruning.
Market Impact Estimation via Causal Foundation Models Introduces a continuous-time causal prior-data fitted network simulator to estimate market impact without confounding biases. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Quantitative finance trading simulation, not AI safety or deployment robustness.
Market Impact Estimation via Causal Foundation Models Introduces a continuous-time causal prior-data fitted network simulator to estimate market impact without confounding biases. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on market impact estimation, not small or efficient language models.
Market Impact Estimation via Causal Foundation Models Introduces a continuous-time causal prior-data fitted network simulator to estimate market impact without confounding biases. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Deals with financial market simulation, not language model evaluation.
Efficient Learning of Survival Analysis Models Under Outcome-Dependent Truncation and Censoring Provides theoretical guarantees for efficiently learning Cox survival analysis models using first-order optimization under outcome-dependent truncation and censoring. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Statistical survival analysis optimization, not LLM evaluation.
Efficient Learning of Survival Analysis Models Under Outcome-Dependent Truncation and Censoring Provides theoretical guarantees for efficiently learning Cox survival analysis models using first-order optimization under outcome-dependent truncation and censoring. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on survival regression, not model compression or efficient serving.
Efficient Learning of Survival Analysis Models Under Outcome-Dependent Truncation and Censoring Provides theoretical guarantees for efficiently learning Cox survival analysis models using first-order optimization under outcome-dependent truncation and censoring. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Addresses statistical survival modeling, not safety or guardrails in deployment.
TACROSS: An Efficient and Low-Cost Scalable Human Touch System Across Heterogeneous Tactile Sensors for Dexterous Robot Learning Develops a low-cost piezoresistive glove system that aligns human touch demonstrations with robot sensors via contact events. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Presents a robotic tactile hardware and learning system, not LLM evaluation.
TACROSS: An Efficient and Low-Cost Scalable Human Touch System Across Heterogeneous Tactile Sensors for Dexterous Robot Learning Develops a low-cost piezoresistive glove system that aligns human touch demonstrations with robot sensors via contact events. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Discusses low-cost tactile hardware, not small language models or inference efficiency.
TACROSS: An Efficient and Low-Cost Scalable Human Touch System Across Heterogeneous Tactile Sensors for Dexterous Robot Learning Develops a low-cost piezoresistive glove system that aligns human touch demonstrations with robot sensors via contact events. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Covers tactile imitation learning, not AI safety or red-teaming.
TACROSS: An Efficient and Low-Cost Scalable Human Touch System Across Heterogeneous Tactile Sensors for Dexterous Robot Learning Develops a low-cost piezoresistive glove system that aligns human touch demonstrations with robot sensors via contact events. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:29	Not relevant Focuses on tactile sensor data transfer for robotics, not computer vision.
Score-Based Learning of Cluster DAGs from Interventions Presents a score-based method to learn cluster causal directed acyclic graphs from interventional data under linear Gaussian assumptions. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:29	Not relevant Statistical causal discovery method, not AI safety or deployment guardrails.
Score-Based Learning of Cluster DAGs from Interventions Presents a score-based method to learn cluster causal directed acyclic graphs from interventional data under linear Gaussian assumptions. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Studies causal graph structure learning, not LLM benchmarks.
Score-Based Learning of Cluster DAGs from Interventions Presents a score-based method to learn cluster causal directed acyclic graphs from interventional data under linear Gaussian assumptions. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Focuses on causal DAG learning, not small or efficient neural models.
Interval-valued SHAP in Tree-Based Models Introduces an imprecise Dirichlet model method to calculate interval-valued Shapley values and analyze attribution robustness in tree-based models. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on Shapley value estimation for tree models, not LLM evaluation.
Interval-valued SHAP in Tree-Based Models Introduces an imprecise Dirichlet model method to calculate interval-valued Shapley values and analyze attribution robustness in tree-based models. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Addresses explainability in tree-based models, not model efficiency or distillation.
Reliability-Aware Future Conditioning for Temporally Robust Robot Manipulation Proposes a reliability-aware conditioning method that mitigates harmful temporal misalignments when using generated video futures for robot manipulation. Preprint	Small and efficient models via arxiv · 12 Oct, 16:29	Not relevant Not focused on small models or computational efficiency.
Reliability-Aware Future Conditioning for Temporally Robust Robot Manipulation Proposes a reliability-aware conditioning method that mitigates harmful temporal misalignments when using generated video futures for robot manipulation. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:29	Not relevant Focuses on robust conditioning for manipulation, not LLM benchmarks.
Reliability-Aware Future Conditioning for Temporally Robust Robot Manipulation Proposes a reliability-aware conditioning method that mitigates harmful temporal misalignments when using generated video futures for robot manipulation. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:29	Not relevant Addresses robot manipulation control under timing shifts, not industrial computer vision.

11,161–11,220 of 13,754← Previous Next →